Introduction toOpendTect

dGB Earth Sciences - OpendTect version 5.0

Prepared for:

Training Manual(July 2015)

Prepared by:

dGB Earth Sciences B.V.Nijverheidstraat 11-27511 JMEnschedeThe Netherlands

Tel: +31 53 4315155Fax:+31 53 4315104

E-mail: info@dgbes.comWeb site: http://www.dgbes.com

- 2 -

Table of Contents1

About this manual 4Licenses 6About F3 Demo dataset 7Support Options 91.Part I: Free Software 121.1 About OpendTect 121.2 Set up a Survey & Load Data 131.2.1   Survey Definition 141.2.2   SEGY Scan Setup & Load 151.2.3   Import Horizon 171.2.4   Import Well Data 181.2.5   Multi-Well Import 20

1.3 Basic Interaction 201.3.1   Tree, Scene & Elements 211.3.2   View & Interactive Mode 231.3.3   Random lines 281.3.4   Save & Restore Session 30

1.4 Attribute Analysis & Cross-plots 311.4.1   Bright Spot Detection and Visualization 341.4.2   Spectral Decomposition 411.4.3   Cross-plots 49

1.5 Seismic Interpretation 601.5.1   Synthetic-to-Seismic Matching 601.5.2   Horizon Tracking 631.5.3   Fault Interpretation 701.5.4   Velocity Gridding & Time-Depth Conversion 73

2. Part II: Commercial Software 822.1 OpendTect Pro 822.2 Commercial Plug-ins 832.2.1   dGB Plug-ins 832.2.2   ARK CLS & Earthworks Plug-ins 872.2.3   ARKeX Plug-ins 892.2.4   Sitfal Plug-ins 89

2.3 Attributes & Filters 892.3.1   Dip-Steering 902.3.2   Attributes for Faults & Fractures 1112.3.3   Ridge Enhancement Filter (REF) 1122.3.4   Frequency Enhancement (Spectral Blueing) 1142.3.5   Flat-Spot Detection 122

- 3 -

2.3.6   Seismic Object Detection Using Neural Networks 1322.4 HorizonCube and Sequence Stratigraphy 1572.4.1   HorizonCube 1592.4.2   Sequence Stratigraphy Interpretation System (SSIS) 1862.4.3   Well Correlation Panel 207

2.5 Seismic Predictions 2102.5.1    Relative Impedance Inversion (SCI) 2112.5.2   Absolute Impedance Inversion (DI & MPSI) 2192.5.3   Porosity Prediction using Neural Networks 229

Appendix - GMT Software 242Personal Notes 245

- 4 -

About this manual

This training manual is prepared for geoscientists who want to learn how to useOpendTect software. The OpendTect suite of software tools consist of a free(open source) seismic interpretation system (hereinafter OpendTect) and a set ofcommercial (closed source) plug-ins. This manual teaches both parts. Theoryand background of some of the more advanced work flows are given but the focusis on hands-on exercises that are executed on “F3 Demo”, a 3D seismic data setfrom offshore The Netherlands. Manual and data set are released free-of-chargefor self-training. The same material is used by dGB Earth Sciences, the companybehind OpendTect, in commercial training classes.

To follow the exercises in this manual you need to install OpendTect and down-load F3 Demo. For details, please follow the instructions on this webpage:http://dgbes.com/index.php/support/tutorials/f3-demo-training .

OpendTect is supported on PC-Linux 32 and 64 bits, Mac-OS/X and PC- Win-dows (Vista, 7, 8 32/64 bits). The latest version of OpendTect + plug-ins can bedownloaded from http://opendtect.org/index.php/download.html . The full instruc-tions for installation can be found via this link: http://opendtect.or-g/rel/doc/User/base/chapter10.4_installation.htm

OpendTect itself runs without license keys but the commercial plug-ins are pro-tected by FlexNet license managing software. Entitlement is stored in FlexNetlicense keys that are checked whenever you run one of the commercial plug-ins.A warning message is given in case you are not entitled to run the software (e.g,when a colleague grabbed the last license before you). F3 Demo is a specialdata set. No license checks are made if you work on this data set. In other wordsall exercises in this manual can be performed without license keys.

- 5 -

The training manual consists of two parts:

Part I – Free Softwarel OpendTect

l Set up a survey & data loadingl Basic interactionl Attribute analysis & cross-plotsl Synthetic-to-Seismic matchingl Interpretation of horizons & faultsl Time-Depth conversion

Part II – Commercial Software*l OpendTect Pro**l Attributes & filters

l Structurally oriented filters & attributes (Dip-Steering)l Frequency enhancement (Spectral Bluing)l Flat spot enhancement (Fluid Contact Finder)l Object detection (Neural Networks)

l HorizonCube & sequence stratigraphyl Global Interpretation (HorizonCube, Well Correlation Panel)l Sequence stratigraphic interpretation (SSIS)

l Seismic Predictionsl Band-limited inversion (Colored Inversion)l Full-bandwidth inversion (Deterministic Inversion)l Stochastic inversion (MPSI)***

*Not all commercial extensions are introduced in this manual. For training of Syn-thRock (stochastic pseudo-well modeling & HitCube inversion) and VelocityModel Building (pre-stack NMO & RMO picking) please contact dGB. For XField(potential field – gravity & EM - modeling) please contact the developers ARKeX.**OpendTect Pro will be released in OpendTect version 6.0 which is due in Q42015.**Multi-Point-Stochastic-Inversion is introduced in this manual. For a more com-plete (commercial)training class please contact dGB.

- 6 -

Licenses

OpendTect is released under a triple licensing policy:

l GNU / GPL license – grants access to the free (open source) part only.l Commercial license – grants access to both parts free (open source) andcommercial (closed source).

l Academic license – grants access to both parts to Universities for educationand R&D only.

Under the GNU / GPL license, OpendTect is completely free-of-charge, includingfor commercial use.

The commercial license permits the user to extend the software with (closedsource) plug-ins. Licenses can either be bought (permanent license) or rented ona monthly or annual basis. The closed source parts of OpendTect are protectedby FlexNet license keys.

Under the academic license agreement Universities can obtain free licenses forOpendTect and the commercial plug-ins for R&D and educational purposes.

For more information please visit the website at:http://www.opendtect.org/index.php/download.html

Please note:l As of OpendTect version 6.0 onward all commercial and academic licenseswill be upgraded to OpendTect Pro licenses. For details, see the OpendTectPro Chapter in Part II of this manual.

l All exercises in this manual can be executed without license managingrestrictions because OpendTect does not check license keys when the soft-ware is used on the F3 Demo dataset.

- 7 -

About F3 Demo dataset

Google maps showing the location of the F3 Demo dataset (brown-filled rect-angle)

F3 is a block in the Dutch sector of the North Sea. The block is covered by 3Dseismic that was acquired to explore for oil and gas in the Upper-Jurassic –Lower Cretaceous strata, which are found below the interval selected for thisdemo set. The upper 1200ms of the demo set consists of reflectors belonging tothe Miocene, Pliocene, and Pleistocene. The large-scale sigmoidal bedding isreadily apparent, and consists of the deposits of a large fluviodeltaic system thatdrained large parts of the Baltic Sea region (Sørensen et al, 1997; Overeem et al,2001).

The deltaic package consists of sand and shale, with an overall high porosity (20-33%). Some carbonate-cemented streaks are present. A number of interesting fea-tures can be observed in this package. The most striking feature is the large-scalesigmoidal bedding, with text-book quality downlap, toplap, onlap, and truncationstructures. Bright spots are also clearly visible, and are caused by biogenic gaspockets. They are not uncommon in this part of the North Sea. Several seismic

- 8 -

facies can be distinguished: transparent, chaotic, linear, shingles. Well logs showthe transparent facies to consist of a rather uniform lithology, which can be eithersand or shale. The chaotic facies likely represents slumped deposits. Theshingles at the base of the clinoforms have been shown to consist of sandy tur-bidites.

The original F3 dataset is rather noisy, to remove the noise, a dip-steered medianfilter with a radius of two traces was applied to the data. The median filtered data(see chapter 5 for information on dip-steered filters) was subsequently inverted toacoustic impedance using the industry standard Strata software. A number of hori-zons were mapped on a loose grid to study the sigmoidal shaped structures.Continuous horizons were created from these coarse grid interpretations by inter-polation with an inverse distance interpolation algorithm. Within the survey, fourvertical wells are present. All wells had sonic and gamma ray logs. Only two wells(F2-1 and F3-2) had density logs. These logs were used to train a neural networkthat was then applied to the other twowells (F3-4 and F6-1) to predict density from sonic and gamma-ray logs. Porosityin all cases was calculated from density using the formula: Porosity = (2.65 –Density) / (2.65 – 1.05).

The F3 Block is available, along with five other datasets, via the Open SeismicRepository the OpendTect website.

References:

Overeem, I, G. J. Weltje, C. Bishop-Kay, and S. B. Kroonenberg (2001) The LateCenozoic Eridanos delta system in the Southern North Sea basin: a climate sig-nal in sediment supply? Basin Research, 13, 293-312.

Sørensen, J.C., Gregersen, U, Breiner, M and Michelsen, O. (1997) High fre-quency sequence stratigraphy of upper Cenozoic deposits. Mar. Petrol. Geol.,14, 99-123.

- 9 -

Support Options

There are many options and ways of getting help with OpendTect’s interactionsand workflows. All available options are listed online at http://www.dgbe-s.com/index.php/support.html .

User documentation

The user documentation is structured in the same way as OpendTect itself. Thereare separate documents for OpendTect and the plug-ins.

All user-documentations can be accessed in multiple ways:

Online as either HTML or PDF, via http://www.dgbes.com/index.php/support.html

Via the softwareThe help menu

The Help Button in each window will automatically pop-up the mostappropriate (sub-)chapter of the user manual.

Tutorials

- 10 -

Via the 'Tutorials' tab on the Support page of the website (http://www.dgbe-s.com/index.php/support.html), you can find four types of self- study assistance:

Self-study OpendTect Course: This manual is an introduction to all aspects ofthe software including most plug-ins. The course includes this free trainingmanual, and the F3 Demo dataset that allows free access to the commercial plug-ins.

Workflow documentation: This document describes various workflows inOpendTect + plug-ins. We describe the purpose, what software is needed(OpendTect only, or OpendTect + one or more plug-ins), and how to do it.

Tutorial videos: Here the user can find different tutorial and webinar videos like:Start new project, Horizon tracking,HorizonCube webinar, SSIS interpretation, Dip steered median filter,Chimney Cube etc…

How-to-Manuals:OpendTect workflows are explained here, the pdf version canbe downloaded (see link below). Different topics are explained: How to visualize

- 11 -

objects in OpendTect, Creating a good SteeringCube, Stratal Slicing, FaultEnhancement Filtering, How- To RGB blending etc

User Mailing List

There is an active User Community. The mailing list users@opendtect.org is forsharing information relevant to OpendTect users. Anyone on this list can send e-mails to all OpendTect users e.g. to pose or answer questions, suggest work-flows, announce innovations etc. Please do not use this mailing list for supportquestions.Support

For support questions please contact OpendTect’s support team at:support@opendtect.org

Social Media

There are OpendTect user groups on Facebook and LinkedIn.

- 12 -

1.Part I: Free Software

1.1 About OpendTect

What is supported?

OpendTect version 5.0 supports all tools you expect to find in a seismic inter-pretation system. Key features include:

l 2D, 3D & Pre-stack seismicl 2D & 3D viewersl Volume rendering & RGB blendingl Seismic attributes & cross-plotsl Spectral decompositionl Movie-style parameter testingl Distributed computingl Rock-physics libraryl Horizon trackersl Faultsl Well-tiel Depth Conversionl Geobodies and … a lot more …

As stated before all functionality listed above is available free-of-charge whenOpendTect is run under the open source GNU/GPL license. This system can beextended with other free software systems and several open source plugins. dGB

- 13 -

developed links to Madagascar and GMT (see below). Several open source plu-gins developed by others are available via the OpendTect website athttp://opendtect.org/index.php/products/free-products/free-3rd-party-plugins .

Madagascar

The Madagascar link integrates OpendTect with Madagascar, an open sourceseismic processing package that is widely used in R&D circles.

Generic Mapping Tool (GMT)

GMT is an open source collection of tools for manipulating Geographic andCartesian data sets and producing encapsulated postscript (eps.) file illustrationsranging from simple x-y plots via contour  maps  to  artificially   illuminated surfaces  and  3-D perspectives views.

1.2 Set up a Survey & Load Data

In this Chapter you will learn how to set up a new survey (a project) and how toload seismic data, horizons and well data using industry-standard file formatssuch as SEGY, LAS and ASCII.

In many oil companies setting up surveys and loading data is done by specialists.Moreover, there are several commercial plug-ins that support easy project setupand data IO to and from SeisWorks/OpenWorks, GeoFrame-IESX and Petrel.

Since not everybody needs (or wants) to know how to do this the hard way it ispossible to skip this entire Chapter. F3 Demo is already set up for OpendTect,hence there is no need to start from scratch. Simply go to the next Chapter to startyour training.

The raw data for our new survey are located in a folder called Raw_Data in theF3 Demo directory.

- 14 -

1.2.1   Survey Definition

What you should know about OpendTect surveys:

l A survey is defined by a 3D grid of Inline, Cross-line numbers and Z samplerate in time, or depth.

l Inline, Cross-line numbers are linked to X,Y rectangular co-ordinates via asingle linear transformation. (OpendTect does NOT support geographic co-ordinate mapping.)

l Surveys can be set-up for 3D seismic only, for 2D seismic only and for 2Dand 3D seismic data.

l The Z dimension (time or depth) as defined in the survey setup determinesthe primary display axis in OpendTect. Data in the other dimension can bevisualized in new (3D visualization) scenes in which the data is transformedon-the-fly using a given velocity field.

l 3D seismic must fall inside the defined survey boundaries.l It is possible to load 3D seismic data sets with varying orientations andsample rates inside one survey. All data sets are mapped onto the definedgrid.

l 2D seismic lines can stick outside the defined survey boundaries. The griddimensions as defined in the survey setup are used in gridding operations,e.g. when creating a 3D horizon from a 2D horizon.

Surveys can be set up in different ways:

1. Manually: You enter the required information such as the specification ofthe 3D grid and the inline, cross-line to X, Y co-ordinate transformation byhand. The latter is specified:a. Either in the form of two linear functions.b. Or, as three points (usually the corner points of the survey outline).

Two of the points must lie on the same inline. OpendTect derives thelinear transformation functions from the specified points.

2. SEGY Scan: SEGY data usually contains inline, cross-line and X,Y co-ordinate information in the trace headers. As not all SEGY data adheres tothe standard (SEGY Revision1) OpendTect supports tools to help you ana-lyze and where needed correct SEGY files.

3. Copy from another survey..4. Set up for 2D only: For 2D seismic surveys OpendTect only requires X, Y

co-ordinates to be correct. You can set it up with a fake inline, cross-linegrid.

5. Using   Commercial   Tools: There   are   commercial   links   toSeisWorks/OpenWorks, GeoGrame-IESX and Petrel.

- 15 -

1.2.2   SEGY Scan Setup & Load

What you should know about SEGY:

l SEGY is the industry standard seismic data format that was defined ori-ginally as a tape format.

l It consists of a Binary Header with general information like data format, tracelength and sample rate, followed by an EBCIDIC Header (descriptive datatyped in by the seismic processor – not very trustworthy) and seismic traces.Each trace consists of two parts: a trace header followed by trace data.

l Not all SEGY files (especially older files) adhere to the standard definition,which is called SEGY – Rev. 1 (Revision 1).

l Before loading SEGY data you must verify that the information OpendTectneeds (inline, cross-line and X,Y co-ordinates) is stored in the trace headerswhere OpendTect expect these.

l If not, you specify where the information is located. If the information is notpresent at all you can create the information using OpendTect’s traceheader manipulation tools.

l If you run into problems, please check this page from the user doc with pos-sible solutions: http://www.opendtect.org/500/doc/od_user-doc/Default.htm#app_c/app_c.htm%3FTocPath%3D_____15

What you should know about SEGY in OpendTect:

l OpendTect can work directly on SEGY data files.l The advantage is that there is no data duplication.l To use this option OpendTect must scan the SEGY file to construct a tablewith inline, cross-line information from the trace headers.

l Current limitation is that scanned SEGY files cannot be pre-loaded inmemory. Pre-loading of SEGY files will be supported from OpendTect ver-sion 6.0 onwards.

EXERCISE 1.2.2

Objective:Set up a new OpendTect survey and load 3D seismic data from SEGY using theSEGY Scan wizard.

Workflow:

- 16 -

1. Select the “Select/Setup...” option under the Survey menu or click

on the Survey icon .

When starting OpendTect for the first time, you are directly

in the Survey Setup & Selection window.

In the Survey Setup & Selection window, all the surveys in

the directory where you are currently working from are lis-

ted.

2. Press to setup a new survey.3. In the next window, name the new survey and select the survey type(2D only, 3D only or both 2D and 3D)4. The survey ranges (or coordinates) can be filled manually, copiedfrom GeoFrame/Petrel, set for 2D only, copied from another survey orcreated by scanning a SEG-Y file, which is the way we will do it in thisexercise. Select "Scan SEG-Y file(s).." in the initial setup options. Thesurvey domain is either time or depth. In the case of depth, units haveto be applied here. Specify Z to be time and click Next

5. Select the Input SEG-Y file from the Raw Data directory.6. Under Manipulate you can inspect and change headers. The win-dow shows a dump of the EBCIDIC header, the binary header and thefirst trace header. You can scroll through the trace headers with theTrc scroller (bottom-right).A plot of trace headers for a number of traces can be made by press-ing the corresponding icon.Optionally binary header information can be changed and trace head-ers can be modified by specifying mathematical formulae in the middlecolumn. Close the window with Cancel.7. Leave the remaining fields as default and press Next.A report of the first 100 traces is given. Press the Display traces icon

to see and QC the first 100 traces. Press Dismiss.8. You are now in the Determine SEG-Y revision window. ModernSEG-Y files are Revision 1 but unfortunately not all SEG-Y files thatare claimed to be Rev-1 adhere to the standard. This is why we needall these tools to examine and possibly overrule header information.

- 17 -

Select "[No]: The file is NOT SEG-Y Rev.1 - treat as legacy (i.e. Rev 0)", then click Next in the Wizard to proceed.9. In the SEG-Y scan window you can optionally overrule the starttime, sampling rate and the SEG-Y co-ordinate scaling. Press Run toscan theentire file.A report is generated in which you find among others inline and cross-line ranges and amplitude ranges with scaling parameters that can beused in case you wish to save the seismic file in 16, or 8-bit format.Press Dismiss and OpendTect will fill in all the parameters it needs inthe Survey Setup window.

If you are unsure of your settings, you can pre-scan the

SEG-Y using the icon. If the report is correct, pro-

ceed, otherwise adjust your settings.10. Click Ok to select the survey you have just setup and OpendTectwill prompt you to load the seismic file that has just been scanned.Click Yes, specify an output file name and press Run to load the seis-mic data in OpendTect.On import completion, you will see an information window givingsome details of the cube. Click Ok to dismiss this, once read.11. Finally, load an inline (from the tree) to check the data.

You will probably want to change the vertical scale. Go to

View and adjust the Z-scale (fit to scene, if desired) and

save this scale as default.

1.2.3   Import Horizon

What you should know about OpendTect horizons:

l There are two kinds:l 2D horizons (from 2D seismic)l 3D horizons (form 3D seismic)

l Each kind has two types:l Geometric gridsl Attribute grids

l Attributes grids are stored as “Surface Data” with the geometric grid to whichthey belong.

- 18 -

EXERCISE 1.2.3

Objective:Import (geometric) horizons from ASCII files.

Workflow:

1. Choose Survey > Import > Horizon > ASCII > Geometry3D then select the  horizon   Input  Ascii   file   (e.g.  \Rawdata\Surface_data\F3-Horizon- FS6.xyt,though you are free to choose any of the horizons).2. Examine the file to determine the header contents and to check details for theFormat Definition. In the Format Definition, you specify which quantity cor-responds to which column in your file. Optionally, Scan Input File and choose ifyou wish to Fillundefined parts.3. Name the Output Horizon and toggle “Display after import”. Press Ok and thehorizon will be imported and displayed (by default, in full).

If you have attributes in your file, you can import them with the Horizon

geometry. By clicking on the when you give a name to the attrib-

ute. In the Format Definition, you also have to specify the cor-

responding column. This attribute will be stored as Horizon Data of the

horizon being imported.

Attributes can also be imported and added to an already existing hori-

zon by choosing Survey > Import > Horizon > Attribute 3D...

1.2.4   Import Well Data

What you should know about well data:

l Wells are defined by a well name and a well track.l Optionally the following information can be added:

l Time-Depth Curves.l Markers.l Logs.

l Time-Depth curves can be modified (stretched and squeezed) in the Well-tie module (synthetic-to-seismic matching module).

- 19 -

l New logs can be created in the Well manager using OpendTect’s Rock-physics library and math & logic manipulations.

EXERCISE 1.2.4

Objective:To load well data from ASCII and LAS files.

Workflow:

1. First import the well track by choosing Survey > Import > Wells > Ascii > Track2. Select the Well track file (e.g /Rawdata / WellInfo /F02-01_welltrack.txt) andexamine it by clicking on the Examine button. Define the Format Definition (col-1:X, col-2: Y, col-3: Z and col-4: MD). The units are in meters.3. Select the Depth to time model file (e.g. /Rawdata/ WellInfo/F02-01_TD_TVDSS.txt) and Examine the file. Define the Format Definition for theDepth to time model (col-1: Depth-m, col-2: TWT-msec). Check the file header.4. Is this checkshot data?: In this case, yes5. Advanced options are optional.6. Name the output well and press Go to import the track file.7. After the well track is loaded display the well in the survey by right clicking Wellin the tree, next click Add and select your well.

8. To import the logs files click on the Manage Well Data icon then Import orSurvey > Import > Wells > Ascii > Logs9. Press the Import button, then select las file (e.g./Rawdata/WellInfo/F02-01_log-s.las), toggle MD and click on Ok.10. When the logs are imported, select which log file you want to display by rightclick on the Well Display > Properties and, in the left- or right-log tab, select thelog to display and the properties.

11. To add markers click on the Manage Well Data icon , and click the Edit

Markers icon . It is also possible to add markers manually. In this exercise, wewill import markers from an existing file by pressing Read New.12. Select the input file (/RawData/ Well Info/F02-01_markers.txt) and Define theFormat Definition (Col-1: MD and col-2: Name).13. Now  select a color for each marker by double-clicking on the appropriate rowin the Color column. When finished, press Ok.14. Show the markers on the well by right clicking on the well in the tree Display >Properties and open the Markers tab. Toggle on the desired markers and set themarker size etc.

- 20 -

Display settings may be changed on all wells simultaneously using

Apply to all wells.

1.2.5   Multi-Well Import

Different options are available to import multiple wells in OpendTect:

l Import > Wells > Bulk...l Well Track: if the track information of more than one well is in thesame file, use this option. The name of the wells need to be in acolumn.

l LAS: allows for the simultaneous loading of several LAS files.l Markers: as for well track.l Time/Depth Model: as for well track.

l Import > Wells > Simple Multi-Well... allows for the import of multiple welltracks following the same principle as for Bulk Well track. In this case, afterreading the file, the well tracks are listed in a table that can be QC-ed andedited prior to actual import.

Important notice.When you have finished with this Chapter do NOT continue inthe survey you have just created but return to the original F3 Demo data set asthis data set is much richer.

1.3 Basic Interaction

This Chapter deals with basic interactions. You will learn how to display seismicdata on in-lines and cross-lines, how to move lines in the 3D scene, how to zoomin and rotate a scene and how to create random lines.

What you should know about the user interface:

l OpendTect supports multiple 3D scenes.l Each scene has its own Z-axis (time, depth, flattened on a single horizon,Wheeler-transformed).

l There are two interaction modes for 3D  scenes: 1) Interact mode for pickingand moving objects, and 2) View mode for zoom, rotate and pan.

l Each Scene has its own tree from which data objects are added to thescene and from where they can be manipulated.

- 21 -

l The tree does NOT show the entire data base of all stored data in thesurvey  as in some other software  systems. The tree shows which elementshave been loaded or processed on-the fly. These elements reside  in memory  and  are  available  for  visualization  and  further manipulations.

l Most display elements (lines, horizons, slices) have multiple layers for com-paring and co-rendering information.

l Data for display can be loaded from stored files, or processed on-the-fly.l Display elements can be manipulated in the 3D scene (right-click menu) orfrom the tree (right-click menu).

l Seismic sections displayed in the 3D scene also be displayed in separate2D viewers.

l Functionality can be accessed via Menus, Icons and short-keys.

1.3.1   Tree, Scene & ElementsThe basic principle of OpendTect is that the user only loads or calculates what isneeded. Only elements that appear in the tree are currently loaded in memory.This has several advantages over having a tree that is showing elements that arestored on disk:

On one single element, multiple stored volumes can be displayed (up to eight).On-the-fly calculated data (available in memory only) can also be displayed. Thisenables the interpreter to test attributes and evaluate their parameters before com-puting the entire volume. This improves results and saves time.

The tree controls what is displayed in the scene. The user can:

l Display stored data (e.g. seismic, horizons, wells, picksets etc…).l Calculate attributes on the fly (and display them).l Create & edit elements (example: create or edit a horizon).

- 22 -

EXERCISE 1.3.1

Objective:Display an inline in the 3D scene.

Workflow:

Show the seismic data at an inline by doing the following:

1.  Click on Inline in the tree and select Add.2.  Right click on the element <right-click>3.  Select Attribute > Stored Cubes> “4 Dip steered median filter”.

- 23 -

1.3.2   View & Interactive Mode

Interactive Mode1. (Re-) position & resize elements2. Make picks (horizons, faults, picksets)

- 24 -

In interactmode elements can be (re)positioned or resized, picks for horizons andpicksets can be created.

View Mode1. Rotate & pan scene2. Move camera position & zoom

In view mode you can rotate and pan the view. You can also move the cameraposition and zoom in and out: “Mov” moves the camera position back and forth

- 25 -

and “Zoom” zooms in and out from a fixed camera position. “Mov” affects the geo-metry and angles of the survey box, Zoom does not.

Use ‘Esc’ to toggle between interact and view modes. The manual has a tabledescribing        the        different        actions        in        different        modes:http://www.opendtect.org/500/doc/od_userdoc/Default.htm#get_started/mouse_cntrls.htm

EXERCISE 1.3.2

Objective:Learn how to zoom, pan, & rotate a 3D scene and how to move a seismic line.

Workflow:

Part 1: zoom, rotate & pan1. To rotate your display, first select the View mode, then left-click and drag the

scene. To pan the scene (move the scene horizontally & vertically) pressthe scroll wheel and drag.

2. Move the camera in and out using the scroll wheel.

Part 2: positioning elementsElements can be repositioned in several ways.

Option1:1. Select the Inline in the tree by clicking on the line-number (425)2. Fill in 250 (the new inline position) line number in the Slice Position toolbar.

Option 2:

- 26 -

1. Go into View mode, rotate the view so you see the inline such that it is dis-played 'from its side' or 'end on' as much as possible in the scene.

2. Go into Interact mode .3. Left-click and drag the inline to a new location... release the left mouse but-

ton when you have reached the desired new location. The new inline num-ber is displayed in the status bar at the base of the OpendTect window.

4. Now you can either:l Click in the empty black space to confirm the new data area, the data willthen be (re-) loaded automatically.

l Right-click on the slice and select Reset Manipulation to undo the draggingof the slice.

Option 3:1. From the tree, right click on the updated inline-number and select Display >

Position option in the pop-up menu list.2. Position the Inline at 195.

Option 4:1. Scrolling: Right-click on an inline and select Display > Position, by pressing

the Scroll button, elements are moved either manually (select ControlManual) or automatically (select Control Auto)

2. Keyboard shortcuts exist to move slice forward/backward with the stepdefined in the box above the tree. To know what are these shortcuts andoptionally change then, follow Utilities > Settings > Keyboard shortcuts. Thedefault shortcuts are the keys X and Y to move forward and backwardrespectively (the inline number must be highlighted for these keys to haveeffect).

3. For fast scrolling use the volume viewer by doing the following:a. In the element tree right-click on Volume and select Add. This will

insert an empty element in the tree.b. Select a stored volume: right-click on <right-click> and choose Select

Attribute > Stored Cubes > 4 Dip steered median filter.c. In interact mode, click and drag an inline/crossline/z-slice, you can

then go quickly through the all volume.

Look at what you have from all sides. Note the different actions when you are inView mode or in Interact mode. Also note that the data values are read out anddisplayed in both methods, these values are displayed at the bottom of thescreen.

Show crossline 1000 in a similar manner.

- 27 -

Show a part of a Z-slice at 700 ms TWT by doing the following:

1. Right-click on Z-slice and select Add.

2. Go to View  mode, rotate the view so you see the Z slice from above.

3. Go to Interact mode4. Make the frame smaller by dragging the green handle points of the frame. (If

the handles are not apparent when you are in Interact mode, click on the rel-evant slice to 'activate' them.)

5. Click in the empty black space to confirm the new data area, the data willthen be (re-) loaded automatically. (Or Reset Manipulation)

6. Position the frame at 700 ms and select the “4 Dip steered median filter”data volume.

- 28 -

1.3.3   Random lines

What you need to know about random lines:

l Random lines can be created in different ways:l Drawn on time-slices or horizons.l Through wells: A random line can be created by connecting the selec-ted wells. By right clicking on the random line in the tree, and

l selecting Create from wells, a dialog box appears with a list of wellsthat can be selected in order to set up the random line path. This

l option is useful for the Well Correlation Panel.l Along Contours: allows the generation of random lines between spe-cified contour range. For this, an interpreted horizon grid will berequired as contours.

l From Existing: This option allows the generation of a random linefrom an existing random line. There is an option available to generatea random line at some distance away from an existing random geo-metry and store it in a new random line geometry.

l From Polygons: allows creating a  random line  definition  from pre-viously created polygons.

l From Table: allows creating a random line in defining its nodes in atable. Each node is defined by its x/y coordinates and Inline/Crosslineinformation.

l Random lines can be optionally saved in the data base.

EXERCISE 1.3.3a

Objective:Create a random line by drawing on a time slice.

Workflow:

- 29 -

1. Add an empty time-slice in the scene:a. Right-click on Random line element from the tree and select New >

Interactive.b. Use the left mouse button to  draw on  the time-slice nodes of random

line. You may use a freehand line or by clicking on the time- slice toinsert nodes. The pop-up window allows you to select when the nodesof the random line are displayed. When you are done, click on OK.

2. You can interactively move the nodes as follows: Select Interactmode , youwill see the end nodes at all corners of the random line element. A node con-sists of a little vertical cylinder and a horizontal plane:

3. Right-clickanywhere onthe randomline, selectDisplay >Insert node –before node1 to create anew node atthat position.

4. The timerange canbe edited bydragging the cylinder of the nodes up and down, the lateral position can beedited by dragging the little plane of each node.

5. Click somewhere outside the survey box to confirm the position, and selectthe data to be loaded on this random line. Loading may be slowed down asdata needs to be retrieved from different inlines and crosslines.

EXERCISE 1.3.3b

Objective:Create a random line through existing wells.

Workflow:

1. Right-click on Random line (Random line > New > FromWells...). Select allfour available wells, change the well's order accordingly, (e.g. F03-4, F03-2,

- 30 -

F02-1, F06-1) and give a name to your newly created Random line thensave it.

1.3.4   Save & Restore SessionUse Survey > Session > Save…/Restore…/Auto load… to restart your inter-pretation at a later moment. The graphic scene(s), elements in the tree(s), currentattribute set and neural network are all saved and restored.

When clicking Auto load, choose Enable and then Use one for this survey. Selectone session amongst the available ones. The session will restore itself auto-matically the next time you start OpendTect.

Elements that contain attributes that were calculated on the fly can only

be restored if the attribute definition is still valid at the time of saving

- 31 -

the session. If not, you will get a warning message stating that the attrib-

ute cannot be restored.

Attribute calculations take time. A Session restore will go much faster if

you retrieve the data from disk instead of recalculating it on the fly. So,

before you save a session think whether you can retrieve the data from

disk (e.g. a horizon attribute can be saved as Horizon data with the par-

ent horizon. The same display can thus be restored much faster if you

save the attribute first and then select it from Horizon data before sav-

ing the session).

1.4 Attribute Analysis & Cross-plots

What are seismic attributes?

Seismic attributes are all the measured, computed or implied quantities obtainedfrom the seismic data. The two main reasons for using seismic attributes:

1. Visualization (qualitative)l To remove extraneous information in the hope of revealing trends orpatterns not visible in the original data.

2. Data integration (quantitative)l To obtain information carriers from different sources that can be integ-rated by statistical methods.

What you should know about Attributes in OpendTect:

1. Attributes can be computed from post-stack 2D and 3D data and from pre-stack data.

2. Attribute definition and computation are two separate steps:l Step 1: Define how to compute the (2D or 3D) attribute in the cor-responding “Attribute Set” window (input > algorithm > parameters >output).

l Step 2: Compute the algorithm either on-the-fly on  the  display ele-ment of choice (inline, crossline, Z-slice, horizon, sub-volume, pick

- 32 -

sets, fault plane, 3D geobody), or in batch mode to create an attributevolume (via the Processing menu).

3. Attributes can be chained (output attribute 1 is input to attribute 2).4. You  can  create your own attributes using chaining, mathematics and

logical manipulations.5. Attribute parameters can be tested in a movie-style manner.6. Attribute time gates (vertical window) and step outs (lateral step in multi-

trace attributes) are specified relative to any evaluation point (x,  y, z) wherethe attribute is to be computed (Step 2, see above).

7. A time-gate of 30ms that is defined as [-10, 20] means the software willextract data from a time-gate between 10ms above the evaluation point to20ms below the evaluation  point.  The  extracted  data is resampled tosample rate defined in the survey.

8. Filters  are  a  just another  group  of attributes, hence  are  treated  as attrib-utes.

What attributes are supported?

Since it is possible to create one’s own attributes using chaining, math & logic thenumber of attributes supported in OpendTect is without limit. To put order in theattribute maze, dGB supports an Attribute Matrix on their website. The Matrixmaps attributes versus application domains and is ordered in attribute classes.

Please note that will find both free and commercial attributes described in the mat-rix. The type column specifies whether the attribute is free (OS for Open Source)or not.

Attribute Matrix. See https://opendtect.org/opendtect-attributes-matrix/

The application domains (organized in columns) are:

- 33 -

l Structurall Stratigraphicl Siliciclasticl Carbonatesl Fluidsl Noisel Others

The attribute classes (rows) are:

l Amplitude basedl Frequency basedl Multi-trace basedl Impedance basedl Dip & azimuth basedl Processing & Filtersl Meta-attributesl HorizonCube & SSISl Pre-stack attributes

The following list shows which attributes are useful for a specific task. The attrib-ute links will bring you to the entry in the matrix.

l Noise reduction: Dip Steered Median Filter, Frequency Filter, Gap Decon-volution

l Frequency enhancement (spectral balancing): Seismic Spectral Blueingl Fault detection: Similarity, Fault enhancement filter, Ridge EnhancementFilter, Curvature, Dip, Variance, Fault Extraction

l Fracture prediction: Curvature, Azimuthal AVO, Fracture, Inversion toAnisotropic Parameters

l Layer thickness estimation: Spectral decomposition, Instantaneous Attrib-utes

l Porosity estimation: Deterministic Inversion, NN Rock Properties Pre-diction

l Net-pay: Seismic Coloured Inversion, Stratal Amplitude, Net-payl HC presence detection: AVO attributes, Frequency Attenuation, Energy((far-near)x far), Sweetness, Common Contour Binning, Seismic FeatureEnhancement

l HC saturation estimation:Gas Chimneys, Three Term Inversion?l Oil vs. Gas prediction:Gas Chimneys, Three Term Inversion, NN Clas-sification, Spectral decomposition

- 34 -

l Predicting Clastic Lithofacies (sand-silt-shale):o Simple: Energy ((far-near)x far), Frequency, Phaseo Advanced:Waveform Segmentation, Volumetric Segmentation,Fingerprint, Deterministic Inversion, NN Rock Properties Prediction

l Predicting Carbonate Lithofacies:Waveform Segmentation, VolumetricSegmentation, Fingerprint, Deterministic Inversion, NN Classification

l Mapping seismic geomorphology: Lithology (see above): Similarity (indic-ates erosional incision), Dip Attributes, Spectral decomposition

1.4.1   Bright Spot Detection and Visualization

EXERCISE 1.4.1

Objective:Isolate an amplitude anomaly (bright-spot) using attribute analysis and visualizethe anomalous body in 3D using volume rendering.

Bright-spot visualized at inline 250

Workflow:

Step 1: Define an Attribute

In order to define an attribute, launch the attribute set window by clicking on the

Attribute 3D icon.

- 35 -

Now define  the Energy attribute (is the  sum of the  amplitudes squared) as fol-lows:

1. Select attribute type Energy.2. Set input data to be the seismic volume “4 Dip steered median filter”.3. Use the default time gate from [–28, +28] ms.4. Provide a name. There is no restriction on the length of the name: it may con-

tain spaces. It is recommended  to  use  a name  that contains all essentialinformation of the attribute. It helps you remember what this attribute does,and prevents having to go back to this attribute window to see the exactdefinition of the attribute.

5. Press Add as new. Not pressing Add as new does not add the attribute, butupdates the (already listed) current attribute to this new definition. Thiswould result in an attribute that does not correspond to its name. Therefore,always remember to press Add as new when you have created a new attrib-ute to add to the list.

6. Press Close on the bottom right, the option Save on Close is activated bydefault.

7. Provide a (new) name for the attribute set like 'My first attributes' pressSelect. This saves the current attribute set and closes the attribute set win-dow.

Attribute Energy

The attribute set window is active: your defined attributes are now available. Asan exercise, try to describe or sketch in a conceptual sense what the attribute you

- 36 -

just created actually calculates. Click directly on help button on the attributeengine or consult the OpendTect Help function to see if you got it right.

Step 2) Display attribute

1. Add an extra layer to inline 250 by right-clicking on the inline number in thetree > Add > Attribute

2. The attributes available are organized in three categories: Stored, Steeringand Attributes (from the active attribute set and calculated on-the-fly). In theAttributes section, select your attribute Energy [-28,+28]ms. To change theselected attribute, right-click on the listed attribute > Select attribute > Attrib-utes, and select your attribute.

Attribute ‘Energy’ clearly discriminates the bright-spot (inline 250, at 530ms)

Step 3) Color-bar

Visualizing the results is almost as  important as the results themselves. There-fore, try different color-bars for your attribute. Each attribute layer has its own col-orbar. The colorbar is displayed by default above the 3D scene and can be added

in the 3D scene with the icon .

- 37 -

1. First select the attribute from the tree, then change the color-bar (try: chim-ney, faults & grey scales)

2. Right-clicking on the color bar, a menu allows you to flip the color bar,change the Ranges/Clipping (to set the scaling values symmetrical aroundzero as shown above),Manage the color bar, etc. The option Set as defaultsets the specific colorbar by default for all the attributes to be displayed thatdo not have a specific colorbar.

Color-bar Manager

In the color-bar manager, you can edit the colors in  double-clicking or right click-ing on the black markers below the histogram. The right-click menu allows tochange but also remove the colors but also to edit the position and the colors ofthe markers.

The color-bar can be continuous or segmented. When changing Segmentationfrom None to fixed, you define the number of segment. Segmented color-bar are

- 38 -

useful when displaying discrete attribute like Neural Network result (one color cor-responding to one class).

Also the transparency can be modified in moving the red line on the histogram. Italters the parts of the spectrum that are displayed. Double-clicking on the lineinserts another moveable node.

The changes you are making are applied in the same time in your scene so youcan actually QC the color-bar edition. The color-bar can be saved with anothername.

Step 4) Evaluate attribute parameters

Now we are going to evaluate the “Time gate” parameter of the energy attribute byinteractively (movie-style) evaluating its parameter settings:

1. Open the Attribute Set window and select the Energy [-28,+28]ms attributedisplayed on inline 250 (Step 2) in the tree.

2. In the Attribute Set window, select this Energy attribute again and press the

'Evaluate attribute' icon .3. Provide the parameter variations as depicted below and on pressing Cal-

culate all intermediate parameter settings are evaluated

Evaluate 'time gate' window

- 39 -

4. When the calculation is completed, use the slider in this window to switchquickly from one display to the next. In this way you can movie-style inspectthe impact of a parameter change on the attribute. Each step corresponds tothe attribute calculated with a different value of the parameter.

5. When a small time gate is chosen, the attribute response is scattered, whilea large time gate gives a smeared attribute response. Choose the time gatesuch that is an optimal balance between the two.

6. On pressing Accept the current parameter setting is selected and the attrib-ute definition in the attribute set window is updated accordingly.

Step 5) Create a seismic output

So far, everything was done in memory. It means that each time you are dis-playing the attribute on a different element, OpendTect has to first calculate it on-the-fly. Calculating the attribute every time is much more time consuming than cal-culating it once for the complete volume (or sub-volume) and then retrieving thestored attribute. Therefore we are now going to calculate  and store the Energyattribute on disk.

1. Click the Create Seismic Output button or go to Processing > CreateSeismic Output > Attribute > Single Attribute > 3D…

2. Select Energy as the Quantity to output.3. Select a sub-volume: Inline range (100 – 290), Crossline range (920 –

1150), Time range (448 – 600)ms.

Volume output window

- 40 -

Step 6) Volume Rendering

The objective of the last step of this exercise is to fully visualize the bright-spot in3D:

1. Right-click on Volume and select Add. It will insert an empty volume in thetree and centrally in the scene.

2. Position the volume: Right-click on <right-click > >Display > Position. It willlaunch a position dialog. Fill in the ranges:a. Inline range: 100 – 290b. Crossline range 920 – 1150c. Time range: 448 – 600

3. Select ‘Energy’ (processed in Step 5) from the window in the Stored cat-egory

4. Change the color-bar to ‘Chimney’.5. In interact mode, you can now left-click and drag to re-position the surfaces

of the volume or just 'scroll through' in this manner to view the contents.

Visualization of the bright-spot using the volume rendering

- 41 -

1.4.2   Spectral Decomposition

What you should know about spectral decomposition:

Spectral decomposition is used to study seismic data at a sub-seismic resolutionor to study attenuation effects caused by hydrocarbons. The method produces acontinuous time-frequency spectra of a seismic trace. It can be done either byusing Fourier Transformation (FFT) or by using Continuous Wavelet Trans-formation (CWT). The details on both methods have been extensively describedin literature. In general, the technique separates the time series into its amplitudeand frequency components. The FFT involves explicit use of windows, which canbe a disadvantage in some cases. The CWT uses a mother wavelet which isextended and compressed for computing the time-frequency spectra. It is equi-valent to a temporal narrow band filtering. Depending upon the purpose, one ofthe algorithms can be selected.

l FFT is used to delineate the stratigraphic/structural information along aninterpreted horizon.

l CWT is preferably used to delineate hydrocarbon attenuations and thick-ness changes along an interpreted horizon.

- 42 -

EXERCISE 1.4.2

Objective:Study paleo-geomorphological features by displaying 3 iso-frequencies sim-ultaneous with color stacking.

Color stacking, also called RGB blending, allows multiple attributes to be com-bined into one display for simultaneous analysis. The combined data can beassessed through brightness and color balance. In many cases RGB displaysshow features with greater clarity and increased detail compared to standard dis-plays.

Workflow:

1. Right click on Horizon on the tree, click on Add color blended…Choosehorizon Demo 1->MFS4. To speed up the exercise, load a sub selection ofthe horizon: inline 200-650; crossline 500-1000.

In the tree, the horizon appears with 4 separate attribute layers. The three lowestattribute layers represent the RGB channels (see color flags next to each layer).Three attributes can thus be blended into a single display.

The fourth attribute is the alpha channel, which can be optionally added. Thealpha channel will make the horizon transparent where the loaded attribute has ahigh value. When using attributes like similarity, this will thus display low valueareas, i.e. faults/fractures.

2. We need to define 3 different attributes that will be loaded to the RGB chan-

nels of the horizon. Open an attribute set , select Frequency then

- 43 -

Spectral Decomposition as attribute. It shows us the different parameters,which will be used to calculate the Spectral decomposition attribute:

l the Input datal the Transform type (algorithm to use, FFT or CWT)l the Time gate.l the Output frequency.

1) Defining the time gate:

Since the extraction of spectral decomposition is done on a horizon, choosing theright time gate is critical. The time gate represents the interval of investigation. If asymmetrical time gate is chosen (e.g. [-28, +28ms]) the attribute will highlight geo-logical features above and below the horizon. When an asymmetrical time gate ischosen (e.g. [-8, 24ms] or [-24, 8ms]) the attribute response will highlight geo-logical features below or above the horizon.

1. We are interested in the paleo-geomorphological features below the hori-zon. Choose your time gate such that it covers these features.

- 44 -

When you display the Horizon MFS-4 at section only, it becomes clear

that the horizon is not picked exactly on the maximum. Compensate for

this fact when defining your time gate.

2) Defining three frequencies

Three different iso-frequencies will be  blended  in the  RGB display. We  willchoose these frequencies such that they represent the low, middle, and high fre-quencies of the seismic bandwidth around the horizon.

1. Load an inline and reduce the Z range such that it just covers the horizoninterval (as shown below). Load the same seismic you are going to use asinput for your Spectral Decomposition attribute, i.e. “4. Dip Steered MedianFilter".

- 45 -

2. Right-click on the seismic attribute displayed in the tree and display the seis-mic bandwidth at target level by selecting Display > Show Amplitude Spec-trum...

3. Now you can choose your low, middle and high frequencies within the amp-litude spectrum. The low frequency can be selected as being the first peak,while the high frequency as the last peak.

3) Defining the three attributes

All the parameters have been tailored and the spectral decomposition attributescan be defined:1. In the attribute set engine, create the first attribute:

a. Select ‘Spectral Decomposition’ as the attribute to define.b. Input data: 4-Dip steered median filter

- 46 -

c. Use the FFT in ‘Transform type’.d. Set your Time Gate as defined in your previous analysise. Output frequency: type in the frequency corresponding to the low

frequency (change the step to an even number if the frequency iseven).

f. Give a name to the new attribute, and press Add as new.

2. In the same manner, create the other two attributes, i.e. for middle and highfrequencies.

3. Click on Ok, optionally give a name to the new attribute set, e.g.: SpectralDecomposition (if Save on OK is selected)

4) Processing & displaying the results using RGB color blending technique

1. Convert your attributes into Horizon Data. Go to Processing--> Create Hori-zon Output... and select both the attribute you wish to output and the horizonon which you wish to output it. For example, output SD_24_[-8,24] onDemo1->MSF4. The horizon attribute is then stored at the horizon and canbe retrieved at any moment.

Saving as Horizon Data is faster than calculating on the fly along a hori-

zon. Moreover, the process is done in batch, so can be preferred sim-

ultaneously for other Horizon Data.

- 47 -

2. Display the three new Horizon Data on Demo1-> MFS4 as follows:

l Spectral Decomposition - Low frequency is displayed as red thechannel (right-click on the text adjacent to the red channel-->Select attribute-->Horizon Data...)

l Spectral Decomposition - Middle frequency is displayed as thegreen channel (right-click on the text adjacent to the green chan-nel...)

l Spectral Decomposition - High frequency is loaded on the bluechannel. (right-click blue channel...)

Time-Gate [-8,24] for Red: Low Frequency, 24Hz Green:Middle Frequency,44Hz Blue: High frequency, 64Hz

3. When blending the three inputs, the results should be similar to the oneshown below.

- 48 -

4. Try to flip the green channel (Right-clicking on the color bar, a menu popsup which allows you to flip the color bar), what do you notice? Do you seeone feature better than the other ones?

5. Which  paleo-geomorphological features can you  interpret? What can youconclude in terms of depositional environments, water depth, litho-facies,and direction of currents?

5) RGB and RGBA

We normally create RGB with three channels; Red, Green and Blue. A fourthattribute (called Alpha channel) can be optionally added to highlight structural fea-tures like faults/fractures.

1. Add ‘Similarity’ to the fourth layer,a. Define ‘Similarity’ as a new attribute (See Similarity definition in the

exercise of the section 4.5, p. 88)b. right click on the fourth element in Demo1 > select attribute > Attrib-

utes 3D > Similarity.

What do you notice? Do you see any structural features (faults, fractures)?

- 49 -

RGB (above) and RGBA (below) displays

1.4.3   Cross-plots

What you should know about cross-plots in OpendTect:

The cross-plot tool in OpendTect creates 2D cross-plots for analyzing rela-tionships between seismic data and well data. Two types of cross-plots are typ-ically analyzed: seismic attributes vs. seismic attributes and seismic attributes vs.well logs. The data points are extracted in a given volume or in a region of intereste.g. by drawing a polygon. The extracted data is displayed in a spreadsheet. Thespreadsheet is then used to manipulate and plot the data.

- 50 -

The cross-plotting tool has several functionalities. These include the following:

l Scattered plotsl Density plots (useful when larger number of data points are selected)l Regression fitl Multi-data selectionl Interactive on-the-fly Geo-body extractionl Creating Probability Density Functions for rock property predictionsl Vertical variograms analysisl Extracting Picksets for Neural Network predictionl ASCII file outputl Quick cross-plot snapshots

EXERCISE 1.4.3

Objective:Analyze the attribute response of the bright-spot amplitude anomaly by cross- plot-ting the iso-frequency attributes of the previous exercise. In the cross-plot domain isolate  clusters of data points (pick sets in  OpendTect jargon) and observe theirspatial distribution by plotting the data points in the 3D scene.

- 51 -

Attribute vs. Attribute cross-plots are useful for identifying relationships

between attributes, for supervised predictions (neural networks), for cre-

ating bodies etc.

Workflow:

1. Load a color blended Demo-6 -> FS8 horizon2. Launch the Attribute Set window and create three attributes:

a. Spectral Decomposition [FFT] – time gate: [-12,12]ms (24Hz, 44hz,64Hz) Names: FFT [-12, 12]ms 24Hz, FFT[-12,12]ms 44Hz, FFT[-12,12]ms 64Hz.

b. Save the attribute set (optional)c. Apply these three attributes on the horizon (red-24Hz, green-44Hz, and

blue-64Hz): right-click on the text adjacent to the red channel--> Select attrib-ute--> Attributes 3D..., repeat for both green and blue.

- 52 -

3. Observe that there is one prominent bright spot, which is a shallow gaspocket. Two more bright spots stand out along this horizon. We will restrictour cross-plot analysis to the largest one. We will create a polygon alongthe layer of bright amplitude to restrict our area of analysis.a. From  the  Tree  Scene  list,  right  click  on  the  element Pick-

Set/Polygon.b. Select New > Polygon …c. Provide the name for this newly added polygon. Call it ‘Shallow

Bright Spot’.d. It will add the polygon sub-item in the PickSet element. Make sure that

this is selected. Now use left mouse button to click on the horizon tooutline a polygon. When you are done, right click on this newly addedpolygon (in the Tree) to Close Polygon. Finally right click again on thepolygon name and click on Save.

A color blended Demo-6  -> FS8  horizon  (spectral decomposition) map. The redcolored polygon outlines the area of cross-plot data extraction.

- 53 -

Defined-area data analysis

4. Go to the menu Analysis > Cross-plot > Attribute Vs Attribute Cross-plot or

click on the icon in the toolbar to launch the attribute cross-plot window.

(Or launch Attribute Set window again and press the cross-plot button )

5. In the cross-plot window, specify the following:a. Input Attributes – Select the attributes that were created in step-2ab. Select location by – Polygon (select Shallow Bright Spot polygon that

you created in the Step 3)c. Further, change the settings of both the Inline & Cross line steps to 1,

Time step to 4ms - to ensure enough extracted data to give a goodanalysis.

d. In Location filter check the Horizon option. Select the horizon i.e.Demo6 -> FS8.

e. In the cross-plot window, click OK to proceed.

6. This will extract scattered data (attributes) along the surface. The pop-upwindow displays the data in a table similar to an Excel sheet (this cross-plot

table can also be saved using the save icon ). Select the following axis tobe cross-plotted.a. X-axis – FFT [-12,12]ms 24Hz select/highlight the  column and press

the buttonb. Y1-axis – FFT[-12,12]ms 44Hz select/highlight the column and

press the button

- 54 -

7. Press Cross-plot    button.8. This will display a scattered cross-plot between the selected attributes. To

have a better feel of the data's behavior, toggle the density plot on by press-

ing button in the cross-plot and set the color table as Rainbow (see fig-ure below).

9. Now in this cross-plot, you see various responses i.e. high frequency andhigh amplitude, low frequency and low amplitude, low frequency and highamplitudes etc (with different slopes).

a. Use polygon tool to select scattered  points.  Change selection

mode from to and reverse in clicking on the icon.

l Toggle interact mode on and select the separation as shown inthe green polygon of the cross-plot figure below.

l Press (and choose option Selected picks) to display theselected scattered data in the scene. In the scene, save thegreen colored displayed picks (as Picksets/Body) by right click-ing over them. Note: Ensure that nothing in the tree is selected(highlighted).

l (Then repeat the same exercise for other response (marked inthe above image as 'Step 10-b')

- 55 -

A density cross-plot of spectral cross-plot along a horizon, which is plottedbetween two iso-frequency responses (24Hz and 44Hz) within a selected poly-gon. Note that there is a clear vertical separation of bright amplitudes at higher fre-quencies. The selected data within the polygons (black/white) can be displayedin the scene that later on can be stored as Pickset/Body.

- 56 -

The scattered data selected in the  previous figure  are displayed as picksets.Note that the separation of frequency highlights two different regions of the brightspot.

The cross-plot has helped to identify the changes in the gas pocket

that are possibly due to differences in saturation/thicknesses. Option-

ally, you can repeat the exercise from step-6 to cross-plot the FFT

24Hz, 44Hz and 64Hz attributes.

EXERCISE 1.4.4

Objective:Analyze relationships between seismic attributes and well logs using cross-plots.

Workflow:

1. Define some seismic attributes in the attribute set window e.g. instant-aneous amplitude, dominant frequency, energy etc, (just using the defaultparameters  is sufficient for  this exercise).  Save the attribute definition andclose the window.

2. Go to the menu Analysis > Cross-plots > Well Logs vs. Attributes or click on

the cross-plot icon  in the OpendTect toolbar. It will launch the attrib-ute/well cross plotting window.

- 57 -

a. In the upper section, select the desired attributes.b. In the middle section, select all available wells and logs.

l Extract between the Truncation and FS4 markers. Optionally youcan extract within a Time range or Depth range (TVDSS). Bydefault the extraction is done in depth but can also be done intime by selecting the option. The extraction step (m or ms) canbe also modified.

l Distance above/below: Default (0,0)c. Log re-sampling method: Nearest sample. If there is no sample at the

extraction position, all the available methods will look at the differentsamples with a time/depth gate centered on the extraction positionand of the size of the defined extraction step. The nearest samplemethod will take the value of the sample the closest to the extractionposition.

l Increase the radius around the well e.g. 25 (this will replicate thelog values within this radius).

l Filter Position: blank (meaning no sub-selection). It is possible torestrict the extraction area for the attributes.

3. Proceed by pressing Ok button.

4. In the pop-up spreadsheet, select any well log as an X-axis (e.g. GR) vs.one/two seismic attributes as Y1/Y2 axis (e.g. Dominant Frequency).

5. Press cross-plot button to plot the selected data . By default, it will plotscattered points of all wells vs. selected attribute(s). You can also create  a cross-plot of individual well from the  cross-plot window by changing thecombo-box:

6. Repeat the same exercise to create log vs log cross-plots by selecting onlylogs as X and Y1/Y2 axis in the step 4.

What you should know about Bayesian Inversion:

- 58 -

Cross-plots can be used to create Probability Density Functions (PDFs) that aresubsequently used in seismic prediction work flows. The framework in whichsuch predictions are made is called Bayesian inversion. The Bayes’ Theorem inan inversion context is summarized by Kemper (2010):

Posterior distribution is proportional to the Prior distribution multiplied

by the Likelihood distribution.

The prior distribution is extracted from the data using the cross-plot tool in a formof Probability density functions (PDFs). The likelihood distribution is a mul-tiplication factor that can be applied to the PDF in order to provide a relativeweighting between several PDF functions.

Typical application workflows are:

l PDFs are created from cross-plots between well logs. The PDFs are fedwith inverted seismic data to predict the desired rock properties.

l If the data set is too large to analyze in the cross-plot domain PDFs are cre-ated on a subset and applied to the full data set.

l The analysis is performed on a small specific region of the survey, e.g.around a well. The derived PDFs are applied to the full data set to seewhether a similar response exists elsewhere.

EXERCISE 1.4.5

Objective:Perform a Bayesian inversion to predict whether a similar bright spot as the onewe have studied exist elsewhere along the same horizon.

Workflow:

1. Start from the cross-plot data extracted in the Attribute vs. Attribute Cross-plot exercise.

2. We will try to forecast the occurrence of the brightest amplitudes, cor-responding to step 10a. To do so, we need to select and remove the pointsthat are outside this sub-selection. Only the desired response shouldremain.

3. Capture the response by computing  the  Probability Density function (PDF)

of the remaining data. Click on the button. Provide an output name.

- 59 -

4. Perform the same operation for the other cloud (step 10b from the same pre-vious section).

5. PDF are sometimes noisy especially when the input dataset is sparse. Usethe PDF manager accessible from Survey > Manage > Probability Density

Function. There you can browse/edit using the icon or double clickingon the PDF. Then you can either rename the quantities ('Names' tab) or editthe values displayed in a table ('Values' tab). You can view the PDF dis-

tribution by clicking the icon in the Values tab. It will appear so:

Smooth the function using the  icon . Press Ok and save thesmoothed PDF with a new name.

6. The application of the Bayesian inversion requires each input attribute to beprocessed and stored as a volume. Create two output cubes (Processing >Create Seismic Output > Attribute > 3D...): one for the 24 Hz and one for the44Hz component. To save processing time, limit the processing range toinline 530 – 750, cross-line 660 – 1040 and Z range 500 – 700ms. Makesure this is done before going further.

7. Launch the Bayesian inversion from Processing > Create seismic output >Bayesian classification.

8. Select both previously created and smoothed PDF (to add additional PDFclick on More).

9. Provide a weight for each, for example 1 for the first and 0.8 for the second.The weight functions can be constant but also variable and input usingvolumes. For instance the weight could be a function of the well distance, orthe vertical distance to the target.

10. Provide the input volume corresponding to each attribute.11. Specify the output you want to receive.

The Bayesian inversion provides several kinds of outputs:

- 60 -

l The “P” is the probability volume associated to each PDF distribution.l The “Classification: Class” returns an integer corresponding to the mostl likely PDF at each sample location.l The “Classification: Confidence” returns the distance between the mostlikely and second most likely PDF distribution.

l The determination strength gives a number related to the relative position inthe most likely position (Histogram count).

12. Display and compare the output volumes on the FS8 horizon. See if othersimilar bright spots can be recognized. You may also want to load thesecubes in a small 3D volume since the processing is volumetric.

1.5 Seismic Interpretation

In this Chapter you will learn basic interpretation tasks such as tying wells, track-ing horizons and interpreting faults.

1.5.1   Synthetic-to-Seismic Matching

Tying a seismic volume to well data is a major task in interpretation projects. It istypically done at the start of a project to determine which seismic events cor-respond to which geologic markers.

We will assume that all data (inputs for the tie) have been prepared already. Theinputs are:

l 3D seismic Volumel an initial waveletl well data (either sonic and density logs, or an impedance log, and geologicmarkers)

l (Seismic horizons are optional)

EXERCISE 1.5.1

Objective:Tie a well to the seismic and extract a deterministic wavelet.

Workflow:

1. Load well F03-4, right-click on it and select Tie Well to Seismic.

- 61 -

2. This will open the Tie Well to Seismic window. Fill the requested fields asshown below.

l Select the stat 120Hz wavelet as the reference wavelet. This waveletapproximates the bandwidth of the seismic data. This can be easilychecked by comparing the Amplitude Spectrum from a seismic section overthe target interval (accessible from right-click in the tree) with the spectrumof the stat120Hz wavelet (press the corresponding icon  in  Manage wavelets).  Moreover,  approximating  seismic bandwidth is also possible by

- 62 -

generating either a “Ricker” type or a “Sync” type of wavelet with a specificcentral frequency. It is also possible to create your own statistical wavelet(just like the default stat 120 Hz wavelet) from the data by pressing theExtract button.

The data is reverse polarity (i.e.: an increase in impedance gives rise

to a trough).

l Select your Density and Velocity (or Sonic) logs with the appropriate units.These fields are pre-filled with logs with correct units.

l If a Depth-Time model has already been loaded, you may choose to usethat one or to start over from scratch.

3. Click on Run to pop up the well tie display panel:

4. Optionally, the current (loaded) time-depth curve can be fine-tuned bystretching and squeezing the curve. It is achieved by picking matchingevents on both the seismic and synthetic traces: activate pick mode with the

icon.

To see which events match, load the already mapped horizons Demo 1 throughto Demo 7. Zoom in (middle-mouse scroll button) and pan (left-mouse click-and-drag) until you have a display to pick matching events.After picking the events, press Apply Changes to reflect the changes. You canUndo only the most recent step.

- 63 -

Tip: As you can undo only the previous step, you may want to save

your intermediate T/D curve by clicking on the icon and exporting

this. You can (re-)import it at any time via the same window or via the

Well Manager.

5. Additional information can be also displayed:

l The estimated (deterministic) wavelet can be viewed and option-ally saved.

l The scaler applied to the seismic is specified.l The Cross-checking parameters can be checked (by pressingl Display Additional Information) and used to get the bestl correlation.l Compute the wavelet between two levels (e.g. start-end of dataand one of the provided markers). Your computation intervalshould be defined regarding the interval of interest.

6. The Save icon  in the main Well Correlation window allows you to save:

l The initial (loaded) and/or estimated wavelet.l The logs, the calculated Acoustic Impedance, the Reflectivity and the Syn-thetic as logs (associated with this specific well) or as seismic cubes. In thecase of saving as a volume, the volume consists of the trace duplicated anumber of (user-defined times) around the well track.

1.5.2   Horizon Tracking

What you should know about horizon tracking in OpendTect:

- 64 -

l Horizons can be (auto-) tracked and drawn in a 3D scene and/or on 2Dl viewers.l The auto-tracker tracks amplitude differences along maxima, minima orzero-crossings. Optionally, the tracker also tracks using similarity and seis-mic dip (only if you have a Dip-Steering license).

l When the tracker is used in the 3D scene it operates on a sub-volume of thedata. The data inside the tracking box is tracked and QC-ed. When done theuser manually moves the tracking box to the next position and extends thetracking.

l The horizon tracker supports three modes of operation:o Volume mode (auto-tracking in 3 dimensions)o Line track mode (auto-tracking along sections)o Line manual (drawing along sections)

l The tracking toolbar offers functionality to change / move the tracking box, tostart the auto-tracker, to QC and edit the tracking results.

EXERCISE 1.5.2a

Objective:Auto-track a horizon in the 3D scene.

Workflow:

1. Add an inline (use the default data, "4 Dip-Steered Median Filter")2. Right-click "Horizon" in the tree and select Track new .... This will launch a

tracking setup window with the following tabs: Mode, Event, Similarity, Prop-erties.

Mode

- 65 -

Choose the tracking mode: Tracking in Volume, Line tracking, or Line manual.

l Tracking in volume is used to auto-track a horizon inside a user definedtracking area (3D volume or sub-volume). The tracker area can be movedand adjusted inside the survey box. This mode is preferred for most horizontracking and will be used in this exercise.

l Line tracking is used to track a horizon on a line (inline or crossline). Thismode gives more control to the interpreter. It is used in difficult areas. Inbetween the seeds, the horizon is tracked along the line. The result is a gridthat needs to be filled by either auto-tracking or interpolation.

l Using the Line manual mode, you manually pick a horizon. The workflow issimilar to line tracking, with the difference that between seeds a line is user-drawn (assisted by interpolation). This mode is used to cross faults, pushthrough noise zones or interpret unconformities.

Line tracking mode and Line manual mode can be used to interpret

horizons on sections inside OpendTect’s 3D scene and on sections

displayed in an OpendTect 2D viewer.

Event

When Tracking in volume or Line tracking is selected, you need to define severaltracking criteria:

l Input data: Select the seismic data on which you are going to track. This canbe the original seismic volume, or a filtered seismic volume (preferred) or anattribute.

- 66 -

In this exercise: Select the Dip steered median filter.

l Event type: Specify the type of event you want to track. The tracker can tracknegative reflectors (Min), positive reflectors (Max), Z-type zero-crossings(0+-) or S-type zero-crossings (0-+). By default, the data selected is thatwhich is loaded and displayed on the line in the scene.

In this exercise: SelectMax.Ensure you are actually interpreting the event type you have selected.

l Search Window: The tracker search in a time window relative to the lasttracked sample. The tracker searches for the chosen event type based onamplitude.

l Threshold type:o Cut-off amplitude: the absolute amplitude is used as the stopping cri-teria for the tracker. When the tracker encounters a value below thisthreshold, it stops tracking. (For a max-event the tracker stops if thevalue is below this threshold value, and for a min-event when it isabove this threshold value). Tip: when pointing your mouse at theevent, the amplitude value is displayed at the bottom of your screen.

o Relative difference: The tracker compares the amplitude of the lasttracked point to the amplitude of the point that is candidate for tracking.If the difference exceeds the chosen percentage, the tracker stopstracking.

In this exercise: Use Relative Difference.

l Steps...: Step-wise tracking results in a better tracked horizons. Good partsof the horizon are tracked first, followed by the more difficult areas. Thetracker will first track the parts of the horizon that have a low difference to theseeds or parts that have a high amplitude. In subsequent steps the trackersettings become less strict. Therefore, the horizon will be of better qualityand needs less editing.

In this exercise: Set subsequent percentage values (incremental: e.g. 1, 2, 5, 10,20), or subsequent amplitude values (decremented e.g. 2500, 2000, 1500, 1000,500).

- 67 -

l If tracking fails: If the tracker cannot find a neighboring point (that complieswith the specified relative difference or cut-off amplitude), it can either stoptracking or extrapolate the horizon.

When the tracker stops tracking before you want it to, adjust the

Threshold value and/or Similarity threshold before choosing the extra-

polate option.)

Similarity

Similarity is a kind of cross-correlation which takes into account both seismic amp-litude and waveform shape information. A part of a trace around the last trackedpoint is compared to all the trace segments on the neighboring trace around thepoints that lie within the Search window (See OpendTect User Documentation formore detail). The number of comparisons is thus controlled by the search window,

- 68 -

while the compare window controls the length of the trace segments. The meas-ure of Similarity between the trace segments lies between 0 and 1. The trackerwill chose the point that has the highest similarity. When the point with the highestsimilarity has a value below the defined threshold, the tracker stops tracking.Tracking with similarity is more accurate, but it takes more time to compute.

Properties

This tab is used to change the display properties of the horizon i.e. color, linestyle, seed shape, seed size etc.

3. After adjusting the parameters in the tracker setup (which can remain openduring tracking), start picking seeds on a displayed inline/crossline.

4. Pick one or more seeds and display the tracking area using the icon.Resize the tracking area by dragging the green anchors on the edges of thecube, but do not exceed your computer's memory size or the size of yourgraphics card memory.

5. Click on the auto-track icon . After the input data is loaded inside thetracking area the horizon is tracked

- 69 -

6. After tracking a part of the 3D-horizon, move the tracking cube to next place:click the top of the tracking cube and drag. A small part of the horizonshould be inside the new position of the tracking area.

Move the tracking cube to next location. When the Cube is at its desired position,click again.

EXERCISE 1.5.2b

Objective:QC and edit the auto-tracked horizon in the 3D scene.

Workflow:When (a part of) a horizon is tracked, the quality can be best checked by using the

display at section only option .

Click on to adjust the area where the edits are made. Then toggle on QC

plane button to show seismic data on an inline/crossline/Z-slice. You canswitch them though by using the drop down option of this button. The horizon willbe displayed as a line on inline and/or crossline display elements and on the QCplane.

- 70 -

You may drag the QC plane over the tracked horizon to QC further

tracking areas. Shift-click on the tracker plane to QC in another dir-

ection or select another orientation in the tracker toolbar.

If the horizon needs editing, there is an in-built Polygonal tool: with the polygontool, you can delete a part of a horizon to be able to re-track it. First select the

polygon selection tool       from the tracker toolbar and circle the area to remove.

Remove the horizon inside the polygon by clicking the remove icon .

To fill the hole again:l Autotrack will track the hole from the edgesl Pick new seeds on the QC-plain and track from seeds onlyl Interpolate (<right-click> on Horizon in tree > Tools > Gridding... There areseveral geometries and algorithms that may be used to grid horizons for dif-fering results. (For full details about this option, see the OpendTect UserDocumentation)

1.5.3   Fault Interpretation

What you should know about faults in OpendTect:

l We distinguish between fault sticks and fault planes.l Fault planes can be mapped directly from line to line (only advisable forlarge faults that can be recognized easily).

l Alternatively, faults sticks are picked and stored in a fault stick set fromwhich fault planes are created by manually grouping the sticks.

l In the current version (5.0) you can pick horizontal and vertical fault sticksbut you cannot combine sticks from different orientations.

EXERCISE 1.5.3a

Objective:Pick a major fault plane in the F3 demo dataset.

Workflow:

- 71 -

Add an inline with default seismic data to the tree and move the line to inline 250where a fault is clearly present throughout almost the entire column.

1. Right click Fault in the tree, select New...2. The fault interpretation toolbar at the bottom becomes active, the edit stick

button is selected by default:

l Pick faults in any order of preference (from top to bottom, middleetc.)

l by clicking on the seismic section.l After picking fault sticks, stick nodes can be dragged by holdingdown the left mouse button over a node and dragging it to a newlocation.

l Individual stick nodes can be removed with Ctrl + left mouseclick.

l Multiple fault sticks can be removed by selecting the polygon

tool and selecting one or more seeds. The stick nodeschange color (to green) when selected and the polygon will dis-

appear. Thereafter immediately click the recycle icon .Deselection of specific sticks within the selection can beachieved using Ctrl+left click. A single node can be selectedusing left click and can thereafter be removed in the same man-ner as above.

l The green backward- and forward arrows allow for undo andredo respectively.

3. After inserting seeds on the current seismic inline, move the inline to a newlocation. For example, set step to 10 and move inline in any direction usingeither the arrows in the Slice Tool or the keyboard shortcuts.

Your previous interpretation is still visible behind or in front of the

moved seismic slice. Often this is considered as a distraction when

interpreting. Thus, right-click the fault (New Fault 1 by default) in the

tree, select Display and choose Only at sections.

4. Proceed to insert seeds on new inline5. After interpretation, the new fault can be saved either in the toolbar or in the

tree

- 72 -

For saving in the toolbar,a. Display fault in full (right click in tree > Display > de-selectOnly at sections).b. Use the polygon tool to select the whole (or part) of the interpreted fault.

Multiple selections/polygons can be drawn - the currently selected

seeds change color.c. SetMove selection to in the first drop-down list, select fault in the second

and Create single new in the third drop-down list in the toolbar.d. Give a name in the empty field.

e. Set color and (optionally)More transfer settings .f. HitGo!

For saving directly in the tree,a. Right-click on the fault.b. Select either Save or Save as… The latter enables you to specify a name

and Replace in tree. The whole interpretation (all seeds) is saved when sav-ing a fault in the tree.

EXERCISE 1.5.3b

Objective:Pick a set of fault sticks and group these into fault planes.

Workflow:

For this exercise, move the inline with default seismic data to inline 450. Severalsmall-scale faults are visible in the bottom-left corner, all steep dipping slightlytoward the East.

1. Interpret 3-5 fault sticks by inserting seeds.When moving to a New Fault, use shift + left mouse click for the first

seed. Alternatively, fault sticks can be drawn with a smooth click-and-

drag movement.

When you are finished with one stick, un-click and move the cursor to

the start position of the  next stick and  repeat the process. OpendTect

will automatically detect that you are drawing a new stick.2. Individual seeds can be moved around and deleted similarly as in the fault

interpretation workflow (see section above). The polygon/recycle and arrow

- 73 -

tools work similarly.3. After interpreting several fault sticks on inline 450, move the slice (e.g. 10

steps) in any direction and draw new fault sticks.Sticks are not connected by a plane, unlike fault interpretations at this

point.4. Similarly to when interpreting faults, it might be beneficial to show the inter-

pretations on the current section only (right click FaultStickSet in tree > Dis-play> Only at sections)

5. Select the appropriate fault sticks:a. Define a Similarity attribute (see Attribute Analysis).b. Add a time-slice to the scene and move the slice such that it intersects

the fault-sticks.c. Apply the Similarity attribute to the time slice. To save time, resize

slice to the area around the interpreted fault sticks.d. Use the polygon tool to select sticks belonging to one fault (note: the

color of selected seeds is green, unselected seeds are violet).

How to convert FaultStickSets into Faults (Fault planes)

6. EitherMove or copy the selected fault stick set. Then set Fault in the seconddrop-down list and choose Create single new. Set name, color and (option-ally)More transfer settings and hitGo!

7. The new fault is Moved or Copied (according to settings) to the Fault sectionin the tree.

8. Repeat selection process for all fault sticks.Faults and FaultStickSets allow for transferring back and forth, with

multiple options (create single new, create new in series, merge with

existing and replace existing).

1.5.4   Velocity Gridding & Time-Depth Conversion

What you should know about Velocity Gridding and TD conversion:

l Time-Depth conversion is performed on-the-fly by transforming the Z-axis ofthe scene using a given velocity field.

l Velocities are given in the form of:

o Velocity volume.o The TD curve of the specified well.o A user-defined linear velocity function.

- 74 -

l 3D velocity volumes can be created in the Volume Builder.l Some advanced options in the Volume Builder are tied to the VelocityModel Builder license (a commercial plugin).

What you will learn in this Chapter:

l How to load a stacking velocity function.l How to grid stacking velocities and create a 3D velocity volume.l How to display the volume on the fly and batch processing.l How to batch-process cubes for depth survey.l How to batch-process horizons for a depth survey.l How to set-up a new depth survey.

EXERCISE 1.5.4a

Objective:Import stacking velocity functions.

Workflow:

1. Go to Survey > Import > Velocity Function > Ascii…. Locate the file Velo-city_functions.txt in the Rawdata directory.

2. Click on Examine to check the file.a. Select the velocity type: Vrms. (See the  User documentation  to  know

more about the different velocity types.)b. The header size is fixed, consisting of 12lines.c. The format is: X-Y-Time-Vrms respectively in column 1, 2, 3, 4. Time

can be either in millisecond or in second, then chose the correct units.3. Give the output velocity name and click Go and Dismiss.

EXERCISE 1.5.4b

Objective:Specify the gridding workflow to grid the stacking velocity functions.

Workflow:

Start the volume builder module accessible from the icon:

1. Select the Velocity gridder step and add it to the Used steps list with themiddle arrow.

2. In the Edit step window, choose Triangulation as algorithm type.

- 75 -

3. Add a Velocity source. Choose Type: Stored Function and select the inputfunction you imported in the previous exercise.

4. Name this step (e.g. Triangulation of Vrms function)5. Store the setup as Grid velocity function6. For this exercise, do not  press Proceed.  Instead  press “Cancel” because

we want to first display the velocity on-the-fly, once the result is Ok, we goback to start the batch processing of the volume.

EXERCISE 1.5.4c

Objective:QC the gridding workflow by applying it to a section where you co-render the seis-mic data and the on-the-fly gridded velocity field. Thereafter create a 3D velocityvolume by gridding the velocity functions in batch mode.

Workflow:

In time scene of time survey:

1. right-click on the element (Inline 425) > Add Volume processing attributeand selectGrid velocity function.

2. To batch process the volume, re-launch the Volume Builder, select the Gridvelocity function, give an output name and press Proceed.

In depth scene of time survey:

1. From the main menu open the depth scene by doing the following: Scenes> New [Depth].

2. A window pops-up asking you to select or create a velocity model.

- 76 -

OpendTect’s volume builder is a general-purpose gridder. It is not

aware that the volume  you created in  the previous step is a velocity

volume. Therefore, you must now first specify that the gridded volume

created before is a Velocity model

a. Click on Create and Select the velocity model.b. Specify the velocity type (RMS).c. Press OK and OpendTect will scan the file to compute the depth

range for the new scene.d. Press OK and a scene (depth scene) pops up.

3. Display any stored volume on the inline 425 in the depth scene. You willhave to use the corresponding Tree in Depth. You will notice that the scenenow shows data in depth, which has been converted from time data usingthe interval velocity you, selected. This is done on the fly.

Conversion of Vrms to Vint:

RMS Velocity can be used for many purposes including T/D conversion, Velocitypicking, etc … but there are other applications requiring Interval velocity instead.

Do the following to convert one velocity type to another:

1. Processing > Create Seismic output > Velocity > Velocity conversion….2. The tagged velocity (Vrms) will be automatically filled as Input velocity

model. (The velocity is tagged when importing the velocity function).3. Fill the Output Cube as Vint (Interval velocity) > Proceed4. Display the new converted interval velocity and compare it with the rms velo-

city.

- 77 -

EXERCISE 1.5.4d

Objective:Use the velocity volume to time-depth convert volumes in batch mode that canthen be used in a depth survey.

Workflow:

Using the new stored velocity model, cubes from the time domain can be pro-cessed and visualized in the depth domain.

1. To create a depth cube, go to Processing > Create Seismic output > Velo-city > Time – depth conversion ….

2. Select the Velocity Model (in this case, Velocity Model (INT)). Click on Editand then scan to go through the amplitude of the input velocity model.

3. Select the Time volume to be converted.4. Define the Depth range and the step. In Volume sub-selection, use inline

425.

In general the depth volume range does not change laterally from the

original cube (thus InL/XL step stays the same) but the depth Z range

can be larger.

5. Give a name to your new depth cube (e.g. D-4_DSMF) and click on Pro-ceed. The volume Dip-Steered Median Filter will be converted in depth andstored in the time survey with a tag “D” (for depth).

- 78 -

6. To display your new depth cube go to the depth scene, right-click on theinline number > Add > Attribute and select the Depth type followed by thecube you just created.

EXERCISE 1.5.4e

Objective:Use the velocity volume to time-depth convert horizons in batch mode that canthen be used in a depth survey.

Workflow:

In order to display the horizons in depth survey we will need to first export themfrom time survey using the velocity model:

1. Survey > Export > Horizons > Ascii 3D…2. Output type: X/Y3. Output Z: Transformed and for Depth conversion, select Velocity4. Velocity cube: Vrms (Z-unit: Meter)

EXERCISE 1.5.4f

Objective:Set up a new depth survey to work with depth converted data (e.g. seismic, hori-zons etc.)

Workflow:

In order to create a new survey:

1. Survey > Select/Setup... (or click on the icon), then click on2. Give a name: F3_Demo_Depth

- 79 -

3. Copy from other survey…choose survey: F3_Demo

4. Change the Z range: 0-2100 step 5m > Ok

- 80 -

5. Press Apply, then click Ok

EXERCISE 1.5.4g

Objective:Import the newly time-depth converted volumes and horizons into the new depthsurvey.

Workflow:

This assumes that we saved the time-depth converted seismic data inOpendTect’s native CBVS (Common Binary Volume Storage) format.

1. Survey > Import > Seismic > CBVS > from file…2. Click on Select… and browse to the location of F3-Demo (Time survey)3. Select the depth volume created before (D-4_DSMF.cbvs)4. Keep the default Use in place (This means that the physical location of the

cube will still remain in time survey)

- 81 -

In the same manner import horizons: 1. Survey > Import > Horizon > Ascii > Geo-metry 3D…

Now display your seismic and horizons in depth survey.

- 82 -

2. Part II: Commercial SoftwareUnder a commercial, or academic license agreement OpendTect, the opensource seismic interpretation platform, can be extended with a range of closedsource extensions. These extensions are protected by FlexNet license keys. Allextensions, also the ones developed by other vendors are licensed through dGB.

For purchase/maintenance fees please contact dGB via info@dgbes.com.

As stated before the exercises in this manual can be executed without licensekeys as OpendTect does not check license keys when the survey you work on isF3 Demo.

2.1 OpendTect ProdGB is in the process of upgrading OpendTect such that the system will be trulycompetitive with other seismic interpretation systems. This solution will belaunched under the name OpendTect Pro as an extension to OpendTect version6.0 that is due for release in Q4 2015. OpendTect v6.0 will offer improved workflows for horizon tracking and fault interpretation.

Among others OpendTect Pro will offer:

l An interactive basemap with mapping functionality;l Two-way connection to Petrel;l PDF-3D plugin for sharing 3D images (see below) andl An accurate ray-tracer.

Commercial licenses of OpendTect will be automatically converted to OpendTectPro licenses at no additional costs.

The PDF-3D plug-in supports the capture of a 3D scene inOpendTect and to save the captured information in PDF format. The 3D PDF filecan then be viewed, rotated, zoomed, and manipulated in Adobe’s free AcrobatReader software that is installed on most computers. PDF-3D thus greatlyimproves communication of complex seismic interpretations. The PDF-3D plug-into OpendTect allows volume sections, horizons, and interpretation features to beembedded within a secure technical report.

- 83 -

2.2 Commercial Plug-insOpendTect supports commercial and non-commercial plug-ins. Commercial plug-ins are available for more specialized and advanced tasks. dGB and 3rd partyvendors ARKCLS, Earthworks, Sitfal and ArkEX provide commercial plug-ins forOpendTect.

Unless you are working on the F3 Demo training data set commercial plug-inrequire FlexNet license-keys. You may wish to contact info@dgbes.com or fill inthis form to request an evaluation license.

Logical sets of plug-ins have been combined into packages for typical G&G tasks.These packages are licensed on a per annum basis. The following packages areavailable:

l Geophysics: Attributes & Filters. This package contains OpendTect andthe following plug-ins: Dip Steering, Neural Networks, Fluid Contact Finder,Seismic Spectral Blueing, Seismic Feature Enhancement, and WorkstationAccess.

l Geology: Sequence Stratigraphy. This package Contains OpendTect andthe following plug-ins: Dip Steering, HorizonCube, SSIS, Well CorrelationPanel, Seismic Spectral Blueing, Neural Networks, CLAS Lite, and Work-station Access.

l Geophysics: Inversion & Rock Properties. This package containsOpendTect and the following plug-ins: Dip Steering, HorizonCube,Deterministic Inversion, Stochastic Inversion, Seismic Coloured Inversion,Seismic Spectral Blueing, Seismic Net Pay, SynthRock, Neural Networks,CLAS Lite, and Workstation Access.

This manual follows a similar sub-division for training the commercial parts of thesoftware:

l Attributes & Filters.l HorizonCube & Sequence Stratigraphy.l Seismic Predictions.

Before starting the training exercises let’s first give short descriptions of the com-mercial plug-ins per software vendor.

2.2.1   dGB Plug-insAs well as creating the open-source OpendTect software itself, dGB EarthSciences also develops closed-source plug-ins for the OpendTect base. See:http://www.dgbes.com/

- 84 -

Dip-Steering

The dip-steering plug-in allows the user to create a (dip-) Steer-ingCube which contains local dip and azimuth information of seismic events atevery sample location. The cube is essential for structure-oriented filtering (akadip-steered filtering), and improves resolution of numerous multi-trace attributes(e.g. Similarity) by honoring and following dipping reflectors. It also featuresunique attributes like Curvature and Dip. Finally, a SteeringCube is an essentialinput to the HorizonCube

HorizonCube

A HorizonCube consists of a dense set of correlated 3D stratigraphicsurfaces. Each horizon represents a (relative) geologic time line. Horizons are cre-ated either in a model-driven way (stratal / proportional slicing, parallel to upper /lower), or in a data-driven way via a unique dip-steered multi-horizon auto-tracker.

HorizonCubes impact all levels of seismic interpretation. They are used for:

l Detailed geologic model building,l Low frequency model building for seismic inversionsl Well correlationl Sequence stratigraphic interpretation system (SSIS).

- 85 -

HorizonCube displays a dense set of auto-tracked horizons.

Well Correlation Panel (WCP)

TheWell Correlation Panel plug-in is used for picking well markersand correlating markers guided by seismic evidence. In combination with the Hori-zonCube, the interpreter can use the slider for detailed seismic-steered cor-relations.

Neural Networks

The Neural Network plug-in supports Supervised and UnsupervisedNeural Networks. The main application of Unsupervised NN is clustering of attrib-utes and/or waveforms for seismic facies analysis. The Supervised approach isused for more advanced seismic facies analysis, to create object "probability"cubes such asTheChimneyCube® and TheFaultCube® and is used for inversion to rock prop-erties (e.g.: porosity, Vshale, Sw etc.).

- 86 -

Sequence Stratigraphic Interpretation System (SSIS)

The SSIS plug-in (Sequence Stratigraphic Interpretation System) isan add-on to the HorizonCube. SSIS supports full sequence stratigraphic ana-lysis, including automated wheeler transforms, systems tracts interpretation andannotations.

Fluid Contact Finder

FCF is a seismic hydrocarbon detection technique where the seismictraces are stacked with respect to the depth of a mapped surface (common con-tour binning). The objective is to detect subtle hydrocarbon related seismic anom-alies and to pin-point gas-water, gas-oil, oil-water contacts.

Velocity Model Building

The VMB plug-in is used to pick up RMO velocities from pre-stackCommon Image Gathers. RMO velocities are used to update the 3D velocitymodel in PSDM workflows. VMB supports picking on semblance gathers and pick-ing of pre-stack events for input to the PSDM- Tomography plug-in. Two VMBmodules are supported: Vertical update and Horizon update. Models are con-structed from combinations of gridded/smoothed RMO velocities, interval velo-cities and 3D body velocities (e.g. Salt body velocity).

SynthRock

- 87 -

The SynthRock plug-in is a forward pseudo-well modeling and prob-abilistic inversion package supporting wedge models, stochastic models, pre-and post-stack synthetic seismograms and cross-matching (HitCube) inversion.

2.2.2   ARK CLS & Earthworks Plug-insARK CLS make the following commercial plug-ins for OpendTect. See: http://ark-cls.com/

Workstation Access

TheWorkstation Access plug-in is used for direct data access to andfrom SeisWorks and GeoFrame-IESX.

Seismic Spectral Blueing

The Seismic Spectral Blueing plug-in is a technique that uses welllog data (sonic and density) to shape the seismic spectrum in order to optimizethe resolution without boosting noise to an unacceptable level.

Seismic Colored Inversion

- 88 -

Seismic Colored Inversion enables rapid band-limited inversion ofseismic data. SCI is rapid, easy to use, inexpensive, robust and does not requireexpert users.

MPSI - Deterministic and Stochastic Inversion

Deterministic inversion (by Earthworks and ARK CLS) includes a 3Dmodel builder for constructing a priory impedance models using well log and seis-mic horizon data; a 2D error grid generation module for providing spatial inversionconstraints and a model-based deterministic inversion module. Even betterdeterministic inversion results can be obtained if the low frequency model is builtin OpendTect’s volume builder using HorizonCube input.

Stochastic inversion includes the MPSI (Multi-Point Stochastic Inversion) ultra-fast stochastic inversion module for generating multiple geo-statistical realizationsand the utilities for processing the multiple realizations to represent the inversionuncertainty for lithology, porosity, saturation or other attributes as probabilitycubes. This plug-in group also requires the purchase of the deterministic inver-sion plug-in group.

Net Pay

The Net Pay plug-in is an add-on to Seismic Coloured Inversion tocompute net pay and net-to-gross from thin and not so thin reservoirs. Net Pay isbased on BP technology.

Seismic Feature Enhancement

- 89 -

The Seismic Feature Enhancement plug-in is a flat-spot utility thatenhances the signal of consistent flat events and reduces the "noise" of the chan-nel reflections.

2.2.3   ARKeX Plug-insARKeX license the following plug-in for OpendTect. See: http://www.arkex.com/

XField 2D

Create 2D/2.5D geological models by integrating potential field datawith seismics and other geophysical datasets in a 3D workspace.

2.2.4   Sitfal Plug-insSitfal provide the following plug-in. See: http://sitfal.com/

CLAS Lite

The CLAS Lite plug-in (by Sitfal) is a petro-physics package that sup-ports log editing and calculation of derived logs such as porosity, saturation,volume of clay and temperature.

2.3 Attributes & FiltersOpendTect’s attribute engine can be extended with various plug-ins that allowcomputation of advanced attributes and filters. As OpendTect’s user interface isbuilt dynamically information about plug-ins (and certain attribute options in the

- 90 -

user interface) are only visible if the plug-in are installed and a valid license key isavailable. (As stated before the latter is no issue when working on F3 Demo.)

In this Chapter you will learn how to:

l Remove random noise (Dip-Steering).l Sharpen edges (Dip-Steering).l Visualize faults (Dip-Steering).l Enhance the vertical resolution (Spectral Bluing).l Enhance amplitude anomalies (Optical Stacking (free), Fluid Contactl Finder, Seismic Feature Enhancement).l Visualize seismic patterns (Neural Networks).l Create a Chimney Cube (Neural Networks).

2.3.1   Dip-SteeringWhat you should know about Dip-Steering:

The dip-steering plug-in allows you to create and use a (Dip-) SteeringCube. ASteeringCube contains at every sample position the dip in the inline- and cross-line directions of the seismic events. These dips can be displayed as overlays onseismic sections. Please note that you should display the cross-line dip on aninline and the inline dip on a cross-line (right-click menu in the tree). In 2D, theSteeringCube contains the apparent dip in the line direction.

The SteeringCube is used for:

l Structurally-oriented filtering (e.g. dip-steered median filter)l Improving multi-trace attributes by extracting attribute input along reflectors(e.g. dip-steered similarity)

l Calculating some unique attributes (e.g. 3D-curvature, and variance of thedip).

l Dip-Steered auto-tracking of single horizons or multi-horizons as is done bythe algorithm that creates HorizonCube.

- 91 -

From a SteeringCube several valuable attributes can be computed. Most of theseattributes, which require SteeringCube are grouped under type Dip insideOpendTect’s attribute set.

For example, OpendTect supports computation of a whole family of  volumecurvature attributes. These attributes are useful in the interpretations of frac-tures, geo-morphological features and drainage patterns. Other attributes that canbe computed from a SteeringCube are:

l The polar dip or true dip: the dip is measured from the horizontal and therange of the dip is always positive and given in usec/m or mm/m.

l The Azimuth of the dip direction is measured in degrees ranging from -180°to +180°. Positive azimuth is defined from the inline in the direction ofincreasing crossline numbers. Azimuth = 0 indicates that the dip is dippingin the direction of increasing cross-line numbers. Azimuth = 90 indicatesthat the dip is dipping in the direction of increasing in-line numbers.

Detailed vs. Background SteeringCube:

In this training, you will be creating several SteeringCubes. The differencesbetween these cubes are in the algorithms used to calculate them and the use offiltering. SteeringCubes called 'Detailed' are unfiltered or gently filtered, whilethose named 'Background' are heavily filtered. Detailed SteeringCubes containdetails such  as dip  associated  faults or sedimentary structures. BackgroundSteeringCubes contain only the structural dip.

- 92 -

Examples of (left to right): Original Seismic (Full stack), Detailed Steering andBackground Steering

These Steering Cubes have distinct applications:

Detailed SteeringCube

l Dip & Azimuth attributesl Curvature attributesl Guide multi trace attributes (Similarity)

Background SteeringCube

l Dip Steered Median Filterl Diffusion and Fault Enhancement Filterl Ridge Enhancement Filter

In OpendTect there are three different algorithms available for creating Steer-ingCubes (e.g. BG Fast Steering, FFT and Event Steering). Coming few exer-cises will be carried out using the BG Fast Steering algorithm (based on the

- 93 -

phase of seismic signal). More information about the SteeringCube can be foundin the dGB Plug-ins Documentation: http://www.opendtect.org/500/doc/dgb_user-doc/Default.htm

EXERCISE 2.3.1a

Objective:Compute a Detailed SteeringCube.

Workflow:

1. Bring up the Create Steering Seismics window via Processing > Dip Steer-ing > 3D > Create.

2. Select the Original Seismic as input and the BG fast steering as Steeringalgorithm (these steering algorithms are explained in more detail in the dGBPlug-ins Documentation).

3. Use the default Calculation stepout 1, 1, 1. The dip is calculated within asmall cube of 3x3x3 samples around each sample.

4. A mild median filter with a stepout of 1, 1, 3 is used to remove outliers in theSteeringCube.

5. Give the SteeringCube a name (e.g BG111_MF113) and click Proceed(optional for the training).

To save time run this process for only 10 in-lines by clicking on select

in front of volume subselection option and specify a small inline range

(e.g. between 420 – 430).

- 94 -

We recommend using the SteeringCube parameters in it's name, sim-

ilar to the Output Steering in the above image.

EXERCISE 2.3.1b

Objective:Compute a Background SteeringCube.

Workflow:

The Background SteeringCube is a filtered (smoothed) version of the DetailedSteeringCube.

1. Bring up the Create Steering Seismics window via Processing > DipSteer-ing > 3D > Filter...

2. Input: A background Steering is simply a detailed SteeringCube, which isfiltered with a fairly large horizontal and vertical step-out. Therefore, selectthe Detailed Steering as input.

3. The filter stepout is inl:5 / crl:5 / sample:5.This median filter is calculated along a horizontal square (of 11x11

samples), while the median filter used for calculating the Detailed

- 95 -

SteeringCube was calculated in a vertical elongated block (of 3x3x7

samples).4. Name the Background SteeringCube and press OK (optional for the

training).

To save time run this process for only 10 in-lines by clicking on select

in front of volume subselection option and specify a small inline range

(e.g. between 420 – 430).

EXERCISE 2.3.1c

Objective:Compute Dip and Azimuth attributes from a SteeringCube.

Workflow:

1. Go to the Attribute engine and select Dip attribute.2. Chose 2 Steering BG Detailed as input.3. Select Polar dip for output.4. Click Add as new and close the attribute engine (optionally save it).

5. Load Horizon Demo 1 -> MFS4 by right clicking on Horizon in the tree >Add….

6. Compute the Polar dip attribute on the horizon: Add > Attribute > Select theone you just created i.e. Polar dip available under the Attribute section.

- 96 -

Alternatively, you may save the Polar Dip as a horizon data and load it on ahorizon (see the section on Spectral Decomposition)

7. Change the color-bar to Grey Scales.8. Save the attribute layer by right clicking on the attribute name (i.e. Polar Dip)

in the tree (i.e. under horizon Demo 1 -> MFS4) > Save As Horizon Data (ifyou have not done so already).

What you should know about Dip-Steered Attributes:

Dip-Steering is a concept in which information is extracted along seismic reflect-ors. It is used in filters (aka Structurally Oriented Filters) and to compute attributeresponses from seismic events that belong to the same stratigraphy.

For example, let consider the calculation of a similarity attribute. This attributecompares two or more trace segments by measuring their distance in a nor-malized Euclidean space. Two identical trace segments will yield an output valueof one, while two completely dissimilar trace segments will return the value zero.In a horizontally layered dataset this will work nicely, but in a dipping environmentthe results will deteriorate. So, instead of comparing two horizontally extractedtrace segments we  should follow  the local dip to find the trace segments thatshould be compared. The process of following the dip from trace to trace is calledSteering. It requires a SteeringCube as an input.

Steering

- 97 -

The dip-steering plug-in forOpendTect supports two different modes of data-driven steering: Central steering and Full steering.In Central steering the dip/azimuth at the evaluation point is followed to find alltrace segments needed to compute the attribute response.In Full steering the dip/azimuth is updated at every trace position.The difference between 'no steering', 'central steering' and 'full steering' is shownin the following figures.

These figures show the 2D equivalent of steering, which in actual fact

is a 3D operation.

The trace segments are defined by the time-gate in ms and the positions specifiedin relative co-ordinates. The extension parameter determines how many tracepairs are used in the computation. This is visualized in the image below.

- 98 -

Definition of trace positions relative to the reference point at (0, 0) (plan view, look-ing down onto Z)

ExtensionWith None specified, only the trace pairs specified in Trace positions are used tocompute the output.Mirror at 90 degrees (not available when inputis 2D data) and Mirror at 180 degrees means that two similarities are computed:for the specified trace pair and for the pair that is obtained by 90or 180 degrees rotation. When using Full block as extension, all possibletrace pairs in the rectangle, defined by inline/cross line step- out, are computed.The attribute returns the statistical property specified in Output statistic. For a fulldescription of the Extension options, see the Similarity attribute entry in theOpendTect User Documentation.

EXERCISE 2.3.1d

Objective:Compute a Similarity attribute with and without dip-steering.

Workflow:

- 99 -

1. Open the Attribute engine and select Similarity2. Keep the default time-gate [-28,+28]3. Select Extension: Full Block4. Keep the default stepout, i.e. inl:0; crl:15. SelectMin forOutput statistics6. Select Steering Full > 3 Steering BG Background7. Give it a name (FS_Similarity) and click Add as new

Steered Similarity attribute creation

8. In a similar way, define a non-steered Similarity by selecting Steering: Nonein this case.

Non-Steered Similarity attribute creation

- 100 -

Display & Investigate:1. Display on a single inline (250) the seismic data (right-click on In- line in the

scene > right-click under 425 as prompted > Select Attribute > Stored Cubes> 1 Original Seismics > select inline number 425 and change it to 250). Alsoload on this inline 250 the steered and non-steered similarity attributes(right-click on inline number 250 > Add > Attribute > …)

2. Change and flip the color-bars of the similarity attributes.3. What is the influence of dip steering?

Inline 250: Non-Steered Similarity (left) and Steered Similarity (right)

What you should know about Dip-Steered Filters:OpendTect supports quite a few filters to enhance data and to reduce noise. Fil-ters applied along the calculated dip and azimuth are called dip-steered filters(aka structurally-oriented filters). In the following exercises you  will constructedge-preserving dip-steered filters that are capable of reducing random noise andenhancing laterally continuous events. The filters collect their input in a curveddisk (= time / depth gate is [0, 0]) that follows the local dip and azimuth. The dipsteered median filter outputs the median value of the input amplitudes at the eval-uation point. (See dGB Plug-in Documentation for more information).

EXERCISE 2.3.1e

Objective:Remove random noise from the seismic data using a Dip-Steered Median Filter(DSMF)

- 101 -

Workflow:

To apply a dip-steered median filter to a seismic data set, you need to para-meterize an attribute of type “Filters” aptly named “Dip steered median filter”:

1. Start the attribute engine, Analysis > Attributes > 3D.2. Select Filters as attribute type.3. Input Data: 1 Original Seismics.4. Set the step-out (for inl, crl) to 1, 1 (the optimal step-out will be evaluated

later).5. Steering Data: 3 Steering BG background. You need to use a background

SteeringCube because you want the structural Dip information.

Dip steered median filter (DSMF) attribute creation

6. Give the attribute a name (e.g. DSMF for Dip Steered Median Filter) andclick on Add as new. Keep this attribute set window open.

7. Now in the scene, add an inline (e.g. 425) and load the Original Seismic(right-click on In-line number > Add > right-click under 425 as prompted >Select Attribute > Stored Cubes > 1 Original Seismics).

8. Afterwards, load the just defined attribute (right-click on inline number, e.g.425 > Add > Attribute > DSMF).

9. In the Attribute Set window, use the attribute evaluation tool to evaluatethe step-out: set the initial value “0-0”, increment “1-1”, and Nr of steps “5”.Press Calculate.

- 102 -

10. Once the computation is done, move the sliders to change the stepout valueand see the impact in the scene.

Use the keyboard arrow keys, so you can focus your eyes on the

scene.11. Assess which step-out is best (removing random noise, but not too much

smearing)? Once chosen, press Accept and close the attribute set window.

EXERCISE 2.3.1f

Objective:See how much noise is removed by the Dip-Steered Median Filter using the math-ematics attribute.

Workflow:

To see how much noise is actually removed by the filter, you can subtract thefiltered seismic data from the input data. We will do this calculation on-the-fly asfollows:

1. Define another new attribute, this time of type Mathematics.2. Define the formula e.g. : filtered – original and press Set.3. Specify the previous attribute definition (i.e. DSMF, at the top of the drop-

down menu list) as input for “filtered” and the stored volume “1 Original Seis-mics” for “original” and call this attribute “Noise”. Add as new.

4. Apply Noise to the same section (e.g. inline 425) to see what the dip-steered median filter has removed. (right-click on inline number 425 > Add >Attribute > Noise...)

- 103 -

Noise attribute creation

- 104 -

Inline 425: Original Seismic (top), DSMF Seismic (middle) and theremoved noise (bottom).

EXERCISE 2.3.1g

- 105 -

Objective:Enhance low-quality seismic data near faults using a Dip-Steered Diffusion Filter(DSDF).

Workflow:

The dip-steered diffusion filter is used to replace low quality  traces by neigh-boring traces of better quality. This migration will be performed using a Similarityattribute.

To apply a dip-steered diffusion filter to a seismic data set you first need to definesimilarity and dip-steered median filter attributes separately. The dip-steered dif-fusion filter is an attribute of type Position:

1. Start the attribute engine.2. Either define a new Similarity attribute or use the attribute set where it is

already defined.3. Now select/use the Position attribute.4. Input seismic: Similarity (attribute or cube previously created).5. Set the step-out (=radius) to 1x1. (The optimal step-out will be evaluated

later).6. Time gate: [0,0]7. Steering: Full > 3 Steering BG background.8. Operator:Max.9. Output attribute: 4 Dip-steered median filter (attribute or cube previously cre-

ated) and Add as new.10. Add an empty attribute to the tree (on the same inline where you previously

displayed the filters).

11. Now use the attribute evaluation tool to evaluate the step-out: initialvalue 0-0, increment 1-1, and 5 slices. Press Calculate

12. Once the computation is done, move the slider to change the stepout valueand see the impact in the scene.

13. Which step-out is best (removing random noise, without creating too manyartefacts?) Accept this best stepout.

14. Appropriately name this Attribute (DSDF) and Add as new in the attributeset.

- 106 -

Dip steered diffusion filter (DSDF) attribute creation

The dip-steered diffusion filter can also be accessed from the “Default

Attribute Set” using the Drawer icon from the row of icons at the

top.

- 107 -

Inline 425: Original Seismic (top) and the DSDF seismic (bottom).

EXERCISE 2.3.1h

Objective:Remove noise and sharpen edges/faults with the Fault Enhancement Filter(FEF).

Workflow:

- 108 -

The fault enhancement filter is a combination of dip-steered median filter and dif-fusion filter, modifying the seismic volume to enhance fault visibility.Based on similarity, the data is smoothed (DSMF) away from the faults andsharpened (DSDF) at the fault location. The filter is released with the software asone of the default attribute sets in two forms:

l Fault Enhancement Filter: all  basic attributes needed  as inputs for the fil-tering are shielded and the user can only control the amount of smoothing(dip-steered median filter) versus sharpening (dip-steered diffusion).

l Fault Enhancement Filter (expert): the full attribute set definition is shownand can be modified.

Fault Enhancement Filter attribute creation

1. Go to the attribute engine.2. Either define new Steered Similarity, Dip steered median filter (DSMF) and

Dip steered diffusion filter (DSDF) attributes or use a previously definedattribute set, where these three attributes are present.

3. Now go to the Mathematics attribute and write a conditional equation suchas FS_Sim > C0 ? DSMF : DSDF and press Set. Choose the three pre-viously defined attributes in front of the three variables (as shown in theattribute creation figure) and enter 0.8 in front of C0. Give this new attribute aname, e.g. FEF and Add as new.

4. Apply the Fault enhancement filter to a small area of the Z slice (1640msbetween inlines 120-400 and crosslines 350-600).

- 109 -

a. Right-click on Z-slice > Add > right-click on Z-slice number 924 >Display > Position

b. Change Z-slice position by entering the required data rangesand press OK.

c. Right-click under Z-slice 1640 as prompted &gt; Select Attribute&gt; Attributes &gt; FEF

Evaluation of constant C0:At a given position, if the similarity is higher than the C0 value, then the DipSteered Median Filtered seismic is used and the Diffusion filtered seismic oth-erwise. Thus C0 is a critical parameter.

5. If the differences are not clear enough, you may want to improve your Faultenhancement attribute by evaluating the constant C0.

6. Open your attribute set and click on to evaluate the constant C07. Evaluate different constants starting with an initial value C0=0.1 to C0=0.9.

Which constant shows the best result? (more faults visible and less noise)8. Click on 'Accept' to save the constant C0.9. Now create one more Dip-steered Similarity attribute in your attribute set

using the “FEF” attribute as input.

Similarity attribute creation using Fault Enhancement Filtered seismic

- 110 -

10. Apply one by one, the FS_Similarity (calculated using the original seismic)and the FEF_Similarity (calculated using the Fault Enhancement Filteredseismic) to the Z-slice at 1640ms and compare the results (see imagebelow).

A comparison between minimum similarity computed from original seismic (left)and fault enhancement filtered seismic (right).

Both the Fault Enhancement Filter and Fault Enhancement Filter

(expert) can also be accessed from the “Default Attribute Set” using the

drawer icon from the row of icons at the top.

- 111 -

2.3.2   Attributes for Faults & Fractures

EXERCISE 2.3.2

Objective:Compute and compare various attributes that pick up faults and fractures.

Workflow:

1. Load Z slice 1640ms between inlines 120-400 and crosslines 350-600 (asdiscussed in the previous exercise 2.3.1g).

2. Define several attributes (attribute set) that highlight faults:

l Similarityl Polar Dipl Curvature (most positive, most negative, curvedness…)l Similarity (steered and non-steered) on original, DSMF, FEFvolumes

3. Display and compare the different attributes. What do they highlight, andwhy? Which attributes are best under what circumstances, and for which pur-pose (fault or fractures)? See examples below.

Minimum similarity time slice (left) and Polar Dip (right)

- 112 -

Most Negative Curvature (left) vs. Most Positive Curvature (right)

2.3.3   Ridge Enhancement Filter (REF)What you should know about the Ridge Enhancement Filter:

The Ridge Enhancement Filter is a post-processing filter for fault attributevolumes, such as Similarity. It sharpens the attribute response such that faults aremore clearly visible.

The ‘Ridge Enhancement Filter’ is delivered with the software as part of thedefault attribute set.

The set calculates similarity attributes at 9 locations surrounding the evaluationpoint. Then it compares the differences between similarity values in 4 horizontaldirections. The direction perpendicular to a fault usually exhibits the largest dif-ference and is therefore output as the Ridge-enhancement attribute. The effect isa sharper outline of the faults.

- 113 -

The default attribute set is based on Similarity. With only minor modi-

fications, this attribute can also increase resolution of other attributes

like curvature, or volumes as fault probability volumes.

EXERCISE 2.3.3

Objective:Compute the Ridge-enhancement filer attribute for improved fault visualizations

Workflow:

1. Open the attribute engine.

2. Open the default attribute set and select Ridge Enhancement filter.3. Select in the newly popped up prompt window, 4 Dip steered median filter

for Input seismic and 3 Steering BG Background for Input Steering.

4. Save it in the attribute set with the default name “Ridge enhancement filter”.

- 114 -

5. Apply this set to the delimited Z slice at 1640ms of previous exercise.6. Compare the results with the attributes of the previous exercise.

A comparison (Z-slice 1640ms) of minimum Similarity (left) and RidgeEnhancement Filter (right): linear features have become sharper afterapplying Ridge Enhancement filtering.

2.3.4   Frequency Enhancement (Spectral Blueing)What you should know about Seismic Spectral Blueing:

Seismic Spectral Blueing (SSB, by ARK CLS) is a technique that uses well datato shape the seismic spectrum, to optimize the resolution without boosting noiseto an unacceptable level.

The workflow is as follows: an Operator is designed for SSB using both the seis-mic and well data. This operator aims to shape the seismic amplitude spectrumsuch that it becomes similar to that of the well reflectivity spectrum. Once the oper-ator has been derived, it is converted to the time domain and simply applied to theseismic volume using a convolution algorithm. As the SSB technique uses bothseismic and well data, one of the main prerequisites of this workflow is to havegood quality well-to-seismic ties.

Typically, the aim is to design an operator at the zone of interest (target). It is there-fore best to identify a time gate for this interval before proceeding ahead with theSSB workflow. Ideally one should use a good interpreted horizon in the targetzone to guide the seismic traces and well data (log traces). In this manner, thevarious gated log traces should have sample values over a similar geology.However, for this training exercise, in our case we will just use a window intervalinstead.

- 115 -

Here is the workflow for how to create and apply these techniques in OpendTect:

1. Seismic: Amplitude-Frequency plot2. Smoothing of seismic mean3. Well: Amplitude-Frequency plot4. Global trend of well plot5. Design operator6. Apply Operator7. Quality Check

EXERCISE 2.3.4

Objective:Increase the vertical resolution of the seismic data with SSB technique.

Workflow:

Step 1) Launch Seismic Spectral Blueing

a. From within OpendTect main window click menu Analysis>Attribute> 3D or

click the icon to pop up the Attribute Set Window (figure below)b. Select Spectral Blueing in the Attribute type list to show the parameters

required for this attributec. Click Select... to the right of the Input Data label and select 1 - Original Seis-

mic

- 116 -

d. Select the “Single Operator” option to create a global SSB operatore. Click on “Analyze and Create ...” to launch the SSB Module.

Step 2) Select Input Data

To use the SSB application to design an operator, it is first necessary to analyzethe seismic and well data spectra. This is achieved by loading some seismictrace data and well log impedance data (time converted).

Selecting Seismic datai. Pop up the "Select Input data" menu item under the Tools menu bar or click

the iconii. Click on Input Seismic tab to select 1-Original Seismiciii. Click Load Seismic to load the default “40 traces”iv. Set the range to Sub (Interval length should be 500ms to 1500 ms long)

- 117 -

Selecting Well data

i. Click on Input Well Logs tab.ii. Click on Load wells, then select the well F3-04 (only one well is used for

this exercise but multiple wells can also be selected). Right-click on the wellto generate an Acoustic Impedance log if it is not available yet.

iii. As is the case for Seismic, set the range to Sub (Interval length should be500ms to 1500 ms  long), then close the "Select Input Data" window.

- 118 -

Step 3) Design Operator

Various parameters exist which allow you to control how the operator is gen-erated. These changes occur in real time so you will be able to see immediatelythe effect of the change you have made.

a. Pop up the Design Controls Dialog by either going to Tools > Design Con-

trols or by clicking the icon.b. Smooth the amplitude-frequency plot of seismic data (seismic mean)

- 119 -

c. Smooth the amplitude-frequency plot of well data. Limit the “High Cut” toseismic Nyquist frequency of 125Hz. Alternatively go to Tools > AdvancedControls > Time Domain (tab) > Log Data Resampling > Interval Source:Seismic.

- 120 -

d. Tweak the parameters (low cut, high cut) of the design operator (imagebelow) such that the residual operator (blue curve) stays 0 in the frequencydomain, with a quick drop on both sides. The effect of the parameter tweak-ing is immediately visible on the other SSB seismic display window that isupdated automatically. Note, for example, the seismic ringing that is intro-duced when the residual operator is not flat in the low frequencies (Low cutparameter in the 0-8Hz range).

e. If the QC operator does not have the desired shape, parameter settings inthe spectral blueing workflow must be changed, until the correct QC oper-ator shape is established

f. Save the operator by giving it a name. This operator is saved as a wave-let.

- 121 -

g. You can optionally save your session as well.

Step 4) Apply Operator

a. Close the SSB module.b. Return to the SSB definition in the Attribute Set window - the newly-defined

operator is selected.c. Name the SSB attribute and Add as new.d. Apply the design operator on inline 425, compare the result with the original

seismic data and, if satisfied, optionally create a volume output (Processing> Create Seismic Output > Attributes > Single Attribute > 3D > Quantity tooutput > SSB attribute…)

- 122 -

Inline 425: Original seismic (above) and seismic after SSB (below)

2.3.5   Flat-Spot DetectionVarious methods have been developed to detect locally  horizontal seismicevents, which do not follow the stratigraphy of the  geological  layers. Theseevents are potentially Direct Hydrocarbon Indicators, since fluid contacts will mostoften be perpendicular to the pressure gradient, regardless of the structural dip.Multiples will most often also not follow the local stratigraphy, and are a false pos-itive for these detection methods, since they will also be enhanced should they behorizontal seismic events.

Optical Stacking

What you should know about Optical Stacking:Optical stacking is released as a free (open source) attribute in OpendTect.Historically  optical  stacking  was  the  first  stacking  method  to  enhance hydro-carbon anomalies that emerged in seismic interpretation software systems.Optical stacking stacks seismic traces that are on either side of a 2D profile. The2D profile should ideally be oriented such that depth contour lines are crossedperpendicular to the line direction. The traces in the stacking direction can be

- 123 -

expected to have similar fluid effects (same hydrocarbon columns) hence subtlehydrocarbon effects are enhanced.

EXERCISE 2.3.5a

Objective:Enhance the amplitude anomaly on a random line with optical stacking.

Workflow:

1. Load a horizon in the scene… right-click on Horizon > Add... and selectDemo 6-->FS8

2. Display a Similarity attribute on this surface. This similarity attribute needsto be defined first if not already done (please follow exercise 2.3.1d). To addthe similarity attribute… Right-click on the horizon name > Add > Attribute.Here select the previously defined similarity attribute (and change the col-orbar to Similarity).

3. Pick a Random-line interactively going through the structure where the amp-litude anomaly was previously seen. Save the new random line.

- 124 -

4. Display the seismic data along the random line “4 Dip steered median filter”,with the horizon displayed at section:

5. In the Attribute Set window, define the Optical stacking attribute: Type:Volume statistics; Time gate: [0, 0]; Shape: Optical stack; Stack stepout: 15(x the bin size of 25m = 375m); Direction: Perpendicular. Apply it along theprofile (remember how to evaluate interactively the stepout).

Discuss the results: What events have been preserved, what events have beenenhanced, why?

Seismic Feature Enhancement

What you should know about Seismic Feature Enhancement:The seismic feature enhancement plug-in by ARK CLS is an improvement of theoptical stack. The concept is similar, with the exception that a few traces in the pro-file direction may also contribute to the stack. Thus one stepout must be selected for the perpendicular direction, and a  second  stepout in  the  line direction. Fur-thermore the plug-in contains a dedicated user interface to interactively choosethe stacking parameters.

- 125 -

EXERCISE 2.3.5b

Objective:Enhance the amplitude anomaly on a random line with Seismic Feature Enhance-ment.

Workflow:

1. Start the plug-in from the Analysis menu: ARK CLS Utilities > SeismicFeature Enhancement. (Note the similarities in the user interface with theSeismic Spectral Blueing plug-in.) Keep this attribute windows open and goto OpendTect’s main window.

2. Display the node positions of the random line: Use the random line you pre-viously created. Select the random line in the tree menu and change themode to interactive. The nodes will become visible. Display their exact loc-ation in a table by right-clicking on the random line name > Display > Pos-ition...

3. In the SFE main window, press the “Select Data” icon  . Choose the inputseismic volume 4 Dip steered median filter. Provide an output volumename. In the Traverse section, input the same node positions as seen in

- 126 -

Step 2 (Or digitize the random line by picking a polygon that you can selectas input for the “Traverse”). Close both the Select Data and Edit Nodes win-dows.

4. Press now the “Aperture design” icon   . Select a similar stepout as in the pre-vious exercise: 375m (normal aperture). Set the traverse parallel aperture sizeeither to 0 (same as optical stack), or leave it to a small value like 75m. Close thewindow.

5. Apply SFE by pressing the “Preview SFE” icon . Look at the Seismic dis-play (zoom forward/backward with Ctrl+Mouse wheel). The seismic display

can be closed and re-opened with the icon.

- 127 -

6. Press “Start” icon  to run the SFE processing and save the result ondisk.

Fluid Contact Finder

What you should know about Fluid Contact Finder:The Fluid Contact Finder plug-in (formerly CCB, or Common Contour Binning)enables detection and accentuation of subtle hydrocarbon-related anomalies andfluid contacts (i.e. Gas-Water Contact, Gas-Oil Contact and Oil-Water Contact).This is essentially a seismic detection workflow that involves stacking of seismictraces that have common depth values along a surface. This workflow is essen-tially designed for structural traps and can be efficiently used for identifying andamplifying DHIs (Direct Hydrocarbon Indicators) such as flat spots.

It is based on the following principles:

1. The seismic traces that penetrate a hydrocarbon bearing reservoir lyingwithin a (anticlinal) structural trap at the same depth (i.e. that lie on the samedepth contour lines) encounter identical hydrocarbon columns.

2. Therefore stacking traces along these contour lines enhances possiblehydrocarbon effects while stratigraphic variations and noise are canceled.

- 128 -

This plug-in necessarily requires a reference horizon along whose common con-tour lines the stacking of seismic traces is done. It outputs a new 3D volume witha stacked trace (result of stacking along common contour lines) replacing theinput traces, and a FCF stack display (2D section displaying stacked traces atvarious contour values) displayed in a 2D viewer. The FCF stack display has twooptions: flattened and unflattened. The latter is more helpful for detecting flat spotswhich are horizontal events in this display. FCF can also be used for pre- stackanalysis and for enhancing 4D anomalies (through the local FCF option)

Ideally FCF should be applied in depth domain but may also be

applied in time domain in case of a structurally simple overburden.

Prospect Identification and Data Preparation

The FCF - Fluid Contact Finder plug-in must be used with precaution since itsapplication to an entire survey would be meaningless. A major preparation step isthe identification of the individual structural traps, which normally will have trap-specific fluid contacts. This is especially important in case of an overall structurethat is separate in individual compartments by non-communicating faults. Iden-tification of compartments can be done by looking at Z-slices or horizon grids ofsimilarity, using the following workflow:

EXERCISE 2.3.5c

Objective:Enhance the amplitude anomaly associated with a structure with Fluid ContactFinder.

Workflow:

Step 1) Delimitation of an area:

1. Load a horizon in the scene... right-click on Horizon> Add and select Demo6-->FS8

2. Display a similarity attribute along this surface. This similarity attributeneeds to be defined first if not already done (please follow exercise 2.3.1d).To add the similarity attribute… Right-click on the horizon name > Add >Attribute. Here select the previously defined similarity attribute (and changethe colorbar to Similarity).

- 129 -

Optionally, you can also display depth contours: right-click on the horizonname > Add > Contour display and select Z-values. The step and range ofthe  contours  can  be  edited  by right-click on Contour  >  Display > Prop-erties... . Click Apply to see the change in the scene.

3. Create a new polygon: In the tree scene right-click on “PickSet/Polygon”,and select New Polygon... Provide a name like "FCF-FS8" and a color, thenClick Ok.

4. In interact mode create a polygon around the structural trap located aroundposition inline 200, crossline 1075 on FS8 by clicking (left mouse button) onthe surface. The picked locations appear as dots. Once the polygon is fin-ished right-click in the tree scene on the polygon. Use the option “ClosePolygon” and “Save” to store your polygon.

To delete a wrongly picked seed, simply use ctrl + left mouse click, i.e.

click on the seed with left mouse button while keeping the ctrl key

pressed.

- 130 -

Step 2) FCF Application:

5. Open the FCF Main window from the corresponding icon in theOpendTect tools toolbar, or via the Analysis Menu.

6. Select the FS8 horizon used to generate the polygon and select the seismicvolume.

7. In the Volume Sub-selection window, select the option “Polygon” and thenchoose the previously created polygon.

8. For the first test, the contour Z division can be left as the entire Z surveyrange. This can be restricted afterward (e.g. to the horizon contour rangescorresponding to only the structure). The  step-size  defines the  contourinterval. All traces that lie within the contour interval are stacked, hence thestep-size controls the bin-size.

9. The Z range around horizon needs to be defined. This is the interval aroundthe horizon that will be used for stacking.

Once this main window is entirely filled, pressing the “Run” button will launch theextraction and clustering of the seismic data. When this operation is finished theFCF Analysis window appears with a chart. This chart presents the number of col-lected seismic traces per bin and the total number of collected traces N:

- 131 -

The FCF analysis window allows launching of the following processes:

l All collected traces of a single bin can be displayed sequentially in a 2Dviewer with the option “Display: Single Z”.

l The stacked traces can be displayed in another 2D viewer. The X-Axis ofthis display represents contour-time or depth of the bin where the traceswere collected. Use option “Display: Stack” to do this. Stacking can be nor-mal (sum of all traces divided by the number of traces per  bin) or weightedby the RMS value of the seismic traces.

l The stacked traces can be stored in a 3D volume were the original seismictraces are replaced by their corresponding stacked trace.

The FCF main window remains open when performing the FCF analysis. Thisallows multiple FCF analyses and simultaneous opening of multiple 2D viewersfor quick comparisons.

- 132 -

Apply local FCF from the same named attribute with a stepout similar to theoptical stacking. Compare  the three methods. Which one is best? In which situ-ations?

2.3.6   Seismic Object Detection Using Neural NetworksWhat you should know about Neural Networks in OpendTect

Neural Networks in OpendTect are used for:

1. Visualizing seismic patterns along horizons and in 3D. This is a qualitativeapproach using an unsupervised Neural Network (aka clustering or seg-mentation).

2. Visualizing seismic objects such as chimneys, faults, salt domes, anom-alies, etc. This is a two-class classification approach using a supervisedNeural Network (aka object detection).

3. Predicting rock properties such as porosity, fluid content, etc. This is a quant-itative approach using a supervised neural network.

4. Predicting lithology classes. This is a multi-class classification approachusing a supervised neural network.

This chapter deals with visualizing patterns and objects. How Neural Networkscan be used e.g. to predict porosity from inverted acoustic impedance and poros-ity well logs is described in the chapter on rock property predictions.

l What you should know about supervised Networks in OpendTect

- 133 -

The supervised network is a fully-connected Multi-Layer Perceptron(MLP) with one hidden layer (i.e one layer between the input node andthe output neurons). The learning algorithm used is back-propagationwith momentum and weight decay. Momentum is used as a filtering ofthe step directions in the gradient decent algorithm, which has a pos-itive effect on training speed. Weight decay is a method to avoid over-fitting when training. Weights are multiplied by a weight decay factor toreduce the weight values, which results in smoother functions withimproved generalization properties. The program sets the number ofnodes in the hidden layer. In practice, supervised training, the user isteaching the network to distinguish between two or more picksets.

l What you should know about unsupervised Networks in OpendTect

The unsupervised Neural Network is the Unsupervised-Vector-Quant-izer (UVQ). This Neural Network is first trained on a representative setof input vectors (attributes extracted at different locations) to find thecluster centers. Each cluster centre is then represented by a vector.Before the network is saved, the software sorts the cluster center vec-tors on similarity. This has the advantage that in the application phasecolours are distributed smoothly over the cluster centers resulting insmoother images which are easier to interpret. In the applicationphase, each seismic input vector is compared to all cluster center vec-tors yielding two possible outputs: Segment (or Class) and Match. Seg-ment is the index of the winning cluster center. Match is a measure ofconfidence between 0 (no confidence) and 1 (input vector and winningcluster vector are identical).

The  unsupervised segmentation  approach reveals areas with  similarseismic responses and is used extensively as an easy-to-use andquick interpretation tool. Clustering can be achieved  using waveformand also  using  multi-trace attributes such as similarity and curvaturein the hope of picking up fracture- density patterns.

More  quantitative  analysis of  UVQ  results  is  possible  with  the aid of (stochastically) modeled pseudo-wells (e.g. de Groot, 1999).

Waveform segmentation

Unsupervised segmentation of data can be done in two modes: horizon-basedand volume-based. The exercise in this section follows the horizon based (or 2D)approach, which is also called waveform segmentation because the input to thenetwork is a waveform (= time-gate) extracted along the horizon. A 3D-

- 134 -

segmentation scheme is very similar. However, be aware that, in 3D, only attrib-utes not directly related to the phase at the sample location should be used. Ifphase sensitive attributes like amplitude are used, the results will look very muchlike the original seismic data.

For waveform segmentation to be successful you need a good reference horizonto work from and preferably a  layer-cake setting. Furthermore, it should be real-ized that due to convolutional effects the results are influenced by variations in theover- and underburden. Variations on the waveform segmentation theme are pos-sible. For example clustering waveforms from near-, mid- and far-stacks incor-porates AVO effects.

OpendTect supports two ways to create a horizon-based unsupervised seg-mentation: The Standard Neural Network method and the so-called “Quick UVQ”method.

EXERCISE 2.3.6a

Objective:Visualize seismic patterns on a mapped horizon using Quick UVQ waveform seg-mentation.

Workflow:

1. In the tree, right-click on Demo1-->MFS4 >Workflows > Quick UVQ (the hori-zon needs to be loaded in the scene).

2. In the pop up window (shown below), specify:l the Input Cube, i.e. the data the waveforms will be extracted from.l the Number of classes (by default 10).l the analysis window length, [-8, +24]ms, defined regarding the selec-ted horizon.

l Optionally the area can be restricted using an already defined poly-gon.

l Optionally the Neural Network can be save. It could be then retrievedand restored at any time from the main Neural Network window.

Press Ok.

The analysis windows is in ms for time surveys and m or ft (depending

on the units set for the survey) for depth surveys.

- 135 -

Input selection and parameter definition

3. A training window pops up (see below) in which the network is shown withthe average match between input vectors and class centers. Training canbe stopped when the average match flattens out around 90%. When sat-isfied, press OK.

Neural Network training showing Average match, Selected/UsedAttribute(s) & Positions in the training set

4. A Neural Network report window with statistics pops up where the NeuralNetwork parameters (upper section) and results (table). In the table are lis-ted the value of each attribute for each class center.

- 136 -

5. Press Display to visualize the class centers.

Class center display (click the highlighted colorbar icon to change the col-orbar used in the display).

6. At the same time,  the trained Neural Network is automatically applied to allpositions along the guiding horizon and outputs Class and Match grids. TheClass and Match results can be saved later as Horizon Data.

The  same  color bar should be used  for the  class attribute applied

along the horizon and for the class center display to ease the inter-

pretation.

The appropriate color bar needs to be segmented with the same

amount of segment as there are classes. The color bar can be edited

by right-click on the color bar > Manage or Survey > Manage > Color

Tables

(a) and (b) show the results of 2D segmentation at horizon MFS4 using atime gate of [-8, +24]ms.

- 137 -

(a) Quick UVQ class grid: 10 Classes - window [-8, +24]ms

Areas with similar seismic response (similar waveforms) are visu-

alized.

(b) Quick UVQmatch grid: 10 Classes - window [-8, +24]ms (blackregions are of low average match)

Match grids not only gives the idea of average match in seismic wave-

form, but also an interpreter can use that grid to interpret subtle geo-

morphological features as represented in above figure.

- 138 -

l What you should know about the standard unsupervised method

The standard method is a flexible workflow that can be applied amongothers to perform: waveform segmentation, clustering of waveformsfrom multiple input volumes, horizon-based clustering of other attrib-utes, or clustering in 3D.

Standard method workflow

The workflow is explained for clustering seismic waveforms (i.e. seismic amp-litudes within a given time interval) that are extracted relative to the horizon. Theuser can play with two input parameters: the extraction time-window and the num-ber of output clusters.

l The time-window is determined by the thickness of the geological interval ofinterest and depends on the seismic phase and bandwidth (with zero-phasedata the start time should be set above the horizon when the horizon marksthe top of the interval). Usually synthetics and/or log displays are used todetermine the optimal time-window.

l The number of output clusters is typically set to 10. A segmentation resultwith 10 segments can be visually re-grouped into smaller numbers bysimply adjusting the color bar (option Segmentation in Manage colortables). The workflow is schematically depicted below.

- 139 -

Workflow for UVQ waveform segmentation

EXERCISE 2.3.6b

Objective:Visualize seismic patterns on a mapped horizon using the standardunsupervised neural network method.

Workflow:

1. Open the Attribute set window . Open  the  default  attribute 

set    :  Unsupervised  Waveform Segmentation. Select theseismic volume 4 Dip steered median filter as input for the attrib-utes.

This default attribute set is designed to extract each

samples within the time window [-100,+200]ms (with a step

equal to the z step of the survey) regarding the application

sample.

2. Start the 3D Neural Network plug-in ( icon or Analysis >Neural Networks > 3D...)

- 140 -

3. Click on Pattern recognition [Picksets]…

Neural Network management

4. In the new pop up window (see below), set the Neural Networkparameters:

- 141 -

Neural Network parameters selection

l Select the Unsupervised Analysis method.l Set the number of output classes to 10 (defaultvalue).

l Select the input attributes to be used, here select asub-set of all available attributes: the samples from -8ms to +24ms.

All the attributes from the active attribute set are selected by

default (the attributes can be unselected by dragging while

pressing the ctrl key). The on-the-fly attributes are listed

first, followed by the list of the stored attributes (their names

are inside [ ]).

- 142 -

l To select the pickset, there are two options:o If a pickset has already been created, it can beretrieved from the list (all picksets are listedthere).

o Using the Create button, generate a randompickset by specifying the number of picks andthe way to extract (in volume, along a horizonor between horizons). Additionally the area/-volume can be restricted

Create a new pickset

Here the pickset consists in 1000 randomly chosen pos-itions along the horizon Demo 1 --> MFS4. Give a name tothis pickset e.g. 1000 Random on Demo1-->MFS4 andclick OK.

Then select this set from the pickset list.

If you would display this pickset you would see something

like in the figure below.

- 143 -

MFS4 horizon with the pickset ‘1000 Random on Demo1-->MFS4’ displayed

Press OK.

5. The specified waveforms are extracted at the random locations.The data is then displayed in a cross-plot spreadsheet for pos-sible further examination.

6. Press Run to start training the network. The Neural Network isfully trained (and should be stopped to prevent overtraining)when  the  average  match  is about 90% (see  figure below).

Neural Network training performance. Training should bestopped when the average match reaches approx. 90%

Store the network by ticking the box. Give a name, e.g.:UVQ_10_[-8,+24]_MFS4.

- 144 -

Press OK to end the training.

7. The lower part of Neural Network management is now filled.Click on Info to see some network statistics and to display theclass centers.

8. Apply the trained Neural Network to the MFS4 horizon by right-clicking on the horizon entry in the tree  (the horizon needs to beloaded in the scene).

Select the Neural Network type: the two outputs are listedand can be loaded.

Loading these attributes takes some processing

time. To save time, save the result directly as

horizon data (Processing > Create Seismic Out-

put > Attributes > Along Horizon > 3D). The hori-

zon data can then be later loaded at the horizon

location (right-click > Add > Horizon Data).

Display the segment attribute the same way as

the previous exercise. Use a similar color bar for

the display of segment and of the waveform at

the class center. Optionally create an appro-

priate color bar with the same amount of colors

as segments, so every segment has its own

color.

The Match output can also be displayed. The Match willshow you where the segmentation is reliable and wherenot.

- 145 -

Object detection: Fluid migration analysis

What you should know about seismic object detection:

In this example we describe how to create a ChimneyCube, i.e. a(gas) chimney“probability” cube.The same workflow can be applied to enhance any seismic object thatyou want to discriminate from its background, e.g. chimney, salt, faults,anomalies etc. Instead of binary object detection, it is also possible toperform multi-class detection following the same principles. Multi-class detection is typically used for seismic facies classification.

This is similar  to  the previous exercise  of UVQ seg-

mentation. Both methods will output seismic facies maps

(or volumes) but whereas the UVQmethod shows areas

(bodies) of similar seismic response that remain to be inter-

preted, the supervised  result reveals areas (bodies) with a

geologic meaning.

The workflow for supervised detection is schematically shown in thefigure below:

- 146 -

Supervised object detection workflow

The workflow to create a ChimneyCube consists of the followingsteps:

1. Define an attribute set.2. Pick examples of chimneys and non-chimneys (i.e. background

response).3. Train the neural network.4. Apply the trained network to the data.

EXERCISE 2.3.6c

Objective:Create a ChimneyCube with the supervised Neural Networkapproach.

Workflow:

Step 1 - Defining the Attribute Set

- 147 -

1. Open the attribute window from the icon (select 3D).2. Open up the default attribute set for NN Chimney Cube via the

icon. Select the input seismic data volume and the steeringdata volume for the various attributes: 4 Dip steered median filteras input seismic volume   and  3  SteeringCube  BG  backgroundas  input  for SteeringCube.

Default attribute set of Chimney Cube

The attributes in your set should capture the difference

between chimneys and background. Visual inspection of

the data shows that chimneys are visible around inline 120

and around inline 690. The chimneys show up as vertical

noise trails.

The seismic response in the chimneys is chaotic, with lowenergies and low trace-to-trace similarity. Thus it makessense to include attributes such as similarity, curvature,energy and variance of the dip (a measure of chaos) intothe attribute set. The vertical nature of chimneys can becaptured by extracting the same attribute in 3 verticalextraction windows: above, at and below  the evaluationpoint. This gives the network a chance to separate vertical

- 148 -

disturbances from localized disturbances.

Step 2 - Picking example locations

To pick chimneys you first have to recognize them. Apart

from being vertical and chaotic noise trails, chimneys often

exhibit circular shapes in cross- sections on similarity-type

Z-slices. Also, chimneys are often associated withseepage-

related features such as mud-volcanoes and pockmarks on

the (paleo-) sea bottom or with nearby high-amplitude

events. The search for chimneys is thus a search through

the dataset in which you use the visualization tools and

interactive attribute analysis methods that are at your dis-

posal to recognize the chimneys that can be picked. This is

of course an interpretation step and as with all inter-

pretations there is an element of bias that will impact the

final result.

Picks are organized in picksets. We will create two sets (figure abelow): one with example locations of a gas chimney and one withexample locations of undisturbed seismic data.

1. Add inline 690 to the tree2. Right click on PickSet/Polygon in the OpendTect elements tree

ad select New PickSet > Empty.3. In the pop up window, provide a name for the pickset you will cre-

ate now, e.g. Chimney Yes.

- 149 -

It is recommanded to have a name descriptive enough to

be able to easily select it later in the Neural Network

design. Like any other object, pickset can be renamed later.

Press Ok.

4. Select the pickset that is now listed in the tree. Provided you are

in interact mode , pick locations for Chimney Yes on theinline.

At the bottom of the screen and in the tree, you will see the

number of picks in your active pickset.

In pick mode you can create picks by left clicking on the

screen. Picks are made and plotted on the plane that is

nearest to the viewpoint. If you want to rotate the view, use

the Scroll buttons along the side of the screen. To delete a

pick select the correct pickset, then press and hold the Ctrl

key and select the pick.

- 150 -

Input picksets Chimney Yes (blue) and Chimney No(green) at inline 690.

5. Save the pickset by right-clicking on your pickset in the tree, andselect Save as. In the pop up window, either select a new namefor your pickset or chose an existing one. When giving a name inthe previous step, an empty pickset has been created: select thispickset.

Optionally several sets can be combined into a single

group. To save multiple picksets in a single group, select

Save from right-clicking at Pickset one level higher in the

tree, and not at the individual pickset.

6. Now create a new pickset called Chimney No and pick more orless the same number of picks in the seismic data where chim-ney is not expected. Save this pickset separately.

7. Repeat the process of picking chimneys and non-chimneys onother lines. Use  for example the Green arrows to scroll inline690 to a new position 10 (20, ..) lines away where moreexamples can be picked.

It is recommended to use the option Display only at sec-

tions when making picks (right-click on the pickset in the

- 151 -

tree). Not using this option would clutter your screen with all

your picks.

To enhance the quality of your picks, change the attribute

(s) you display from time to time. This gives you the oppor-

tunity to pick on the best example locations for several dif-

ferent attributes, and gives a better diversity of input to the

Neural Network at a later stage.

Step 3 - Training and Viewing a Neural Network

1. Open the neural network window by clicking on the neural net-

work icon and chose the Pattern recognition [PickSets]…option.

If you have many picks and want to use many attributes,

you can select to store the extracted data during the Neural

Network process. This prevents having to re-calculate

everything when you only want to take a few attributes out

of your neural network. In that case you would select the

Stored Network option.

2. In the Neural Network design window (below), the Supervisedand Extract now options are ticked by default. We keep this

- 152 -

selection.

On the input attribute panel, all the attributes defined in theattribute set window are listed and automatically selected:unselect the attributes beginning with “NoNNInput”.

Supervised neural network

In the Output nodes panel, select the two picksets Chim-ney Yes and Chimney No.

Finally, define the Percentage used for test set: 30%.

The data set is split randomly into a training set and a per-

centage for testing. The training vectors are passed through

the network and the error is used to update the weights.

The test vectors are also passed through the network but

the error is only used to check the performance during train-

ing and to avoid overfitting.

Press OK.

- 153 -

3. The data are extracted and are listed in a crossplot table. In thespreadsheet you can analyze and edit the attributes by cross-plotting them against each other and (more interesting) againstthe targets Chimney Yes and Chimney No.

Possibly not all picked locations are valid. For example,

near the edge of the data cube, steering data are not avail-

able because of the trace step-out that was used. Steered

attributes cannot be calculated there, so picks in that area

will not have all attributes defined, and these picks will

therefore be ignored.

Press Run.

4. The Neural Network training windows pops up (figure below).

Training performance of the network

Training is ideally stopped when the error on the test set isminimal. This is the point where the network has theoptimal generalization capabilities (picture below).

- 154 -

Watch the performance, and press  Pause when you are satisfied.Your network is trained when the misclassification does not decreaseanymore, and the RMS error is minimal for both train and test set.

You may press clear and start the training again for

example if you overtrained your network.

The colors of the input attributes change during training.

The colors reflect the weights attached to each input node

and range from white via yellow to red. Red nodes have

more weights attached and are thus more important to the

network for classifying the data. Colors are very useful to

tune a network and throw out attributes that may take up a

lot of CPU time without contributing to the final result.

Optionally you can decide to save your Neural Network. Press OK toend the Neural Network training module.

- 155 -

5. Back in the Neural Network manager, you can decide to Store tothe trained neural network (if not done previously) and to accessadditional Info on the trained Neural Network.

Close the Neural Network manager.

The network have output nodes: Chimney Yes and Chim-

ney No. The values of these outputs will be between

approx. 0 and 1 (after format scaling). The outputs will also

be mirror images of each other: if a position belongs to a

chimney the output of Chimney Yes will be high while that

of Chimney No will be low. When we apply the trained net-

work it will thus be sufficient to output only the Chimney

Yes response. The closer the value to 1 the more ‘prob-

able’ it is that the value belongs to the chimney class. This

type of output makes sense only if we are dealing with a

two-class problem. With more than two classes (e.g. for

seismic facies classification) the option “Classification” is

selected. The output of the Neural Network is then replaced

by two new output nodes: Classification and Confidence.

The former gives the index of the winning class while the

latter is a measure of how confident we are in the result in a

range of 0-1 (0 is no confidence, 1 is very confident).

6. You can now test the training performance on your data, e.g. on(part of) inline 690. Right-click on the element in the tree and addattribute: select the Neural Network category and select the out-put Chimney Yes.

The result may look like the figure below. The “ChimneyYes“ output is displayed using the chimney color bar(between 0.8 and 1.2).

- 156 -

Chimney result on Inline 690: filtered seismic

Chimney result on Inline 690: filtered seismic with pre-dicted chimney overlay.

7. If you are not satisfied with the results, you can change the loc-ation of the picks using the misclassified picksets to enhance theperformance of the neural network.

- 157 -

Output Misclassified Chimney (red). The Misclassified pick-sets can be used to enhance the performance of theNeural Network.

When you are satisfied with the performance of yourNeural Network, you want to apply it to the entire volume tobe able to analyze the results.

Step 4 - Apply the trained network to create the ChimneyCube.

1. Go to Processing > Create Seismic Output > Attribute > SingleAttribute > 3D and select the Chimney Yes as quantity to output.Give a name and click OK.

OpendTect allows you  to process the data  on  multiple

machines. For details, see Batch Processing in the

OpendTect User Documentation.

Optionally you may want to display the volume in 3D usingvolume rendering (see exercise 1.5.1, section 6). This isvery consuming in term of memory so you need to first tryon a small area of interest.

2.4 HorizonCube and Sequence Stratigraphy

The HorizonCube is a step-change technology that opens the door to drasticimprovements in understanding the geological meaning contained in seismic

- 158 -

data: 3D sequence stratigraphy, seismic geomorphology with data driven stratalslicing, improved geologic models, wells correlation, low frequency modeling forbetter seismic inversion etc.

Today, seismic interpreters can look forward to the following benefits:

l Low Frequency Model Building & More Accurate, Robust GeologicalModels

In standard inversion workflows, the low-frequency model is considered the weak-est link. Highly accurate low frequency models can be created by utilizing all thehorizons of the HorizonCube, allowing a detailed initial model to be built.In a similar fashion rock properties can be modeled. Instead of using only a fewhorizons all horizons of the HorizonCube are used, resulting in greatly improvedrock property models.

l Rock Property PredictionsThe highly accurate low frequency models can be used to create geologically cor-rect Acoustic Impedance (AI) and Elastic Impedance (EI) cubes usingOpendTect’s Deterministic and Stochastic Inversion plug-ins. To complete theworkflow, the Neural Networks plug-in is used to predict rock properties from theAcoustic Impedance volume, avoiding the use of oversimplified linear modelswhich cannot accurately describe most rock property relations.These advanced tools bring a high degree of precision to traditional seismic work-flows, resulting in better seismic predictions and more accurate input into thereservoir management decision-making process.

l Sequence Stratigraphy (SSIS plug-in)The SSIS plug-in works on top of the HorizonCube plug-in. Users can inter-actively reconstruct the depositional history in geological time using the Hori-zonCube slider, flatten seismic data in the Wheeler domain, and make full systemtracts interpretations with automatic stratigraphic surfaces identification and base-level reconstruction.

l Well Correlation (WCP plug-in)The Well Correlation Panel plug-in is an interactive tool for correlating well dataand for picking well log markers in a consistent manner. The tool supports dis-playing and manipulating multiple wells with logs, markers, and stratigraphiccolumns, plus the connecting seismic data (2D lines, or Random lines from 3Dvolumes) with interpreted horizons, faults, HorizonCube and interpreted systemstracts.

- 159 -

HorizonCube Applications

In this Chapter you will learn how to:

l Create data-driven and model-driven HorizonCubes.l Truncate HorizonCubes.l Extract horizons from a HorizonCube.l Track single horizons from a Steering Cube.l Correlate between wells.l Wheeler transform data (= flattening).l Interpret systems tracts.

2.4.1   HorizonCubeWhat you should know about HorizonCubes:

l HorizonCubes consist of a dense set of (dip-steer) auto-tracked, or modeledhorizons.

l HorizonCubes exists for both 3D and 2D seismic data sets.l Horizons are first order approximations of geologic time lines.l Horizons can never cross each other.l There are two types of HorizonCubes: Continuous and Truncated.l In continuous HorizonCubes all horizons exist everywhere; when horizonsconverge the density of the horizons increases. This tends to happen alongunconformities and condensed sections.

l In truncated HorizonCubes horizons stop when they get too close together.l Using HorizonCube density it is possible to convert continuous Hori-zonCubes to truncated HorizonCubes.

- 160 -

l Flattening on horizons in a HorizonCube is called a Wheeler transform.l Depositional trends and systems tracts are easier to interpret in a Wheeler-transformed, truncated HorizonCube. Model building (interpolating well logsguided by horizons) is easier in a continuous HorizonCube.

l HorizonCube sliders are used in OpendTect to:o Analyze the depositional history.o Identify and extract horizons from the dense set of horizons.o Extract 3D bodies from iso-pach thicknesses or attribute responses.

Details:In standard seismic interpretation workflows, a coarse 3D structural or sequencestratigraphic model of the sub-surface is constructed from a limited set of mappedhorizons. The number is limited because mapping horizons with conventionalauto-trackers, based on tracking amplitudes and similarities, is a time consumingpractice. In particular, mapping unconformities - primary targets in sequence strati-graphic interpretations - is cumbersome with conventional trackers, as amplitudestend to change laterally along such surfaces. HorizonCube maximizes theamount of information that can be extracted from seismic data by significantlyincreasing the number of mapped horizons (figures below).

Seismic section to illustrate the difference between two trackers: conventional vs.dip-steered: (A) Conventionally tracked event based on seismic amplitude andwaveform similarity, (B) the same event has been tracked using the dip-azimuthvolume (SteeringCube).

A HorizonCube consists of a dense set of auto-tracked seismic horizons. Theauto-tracker tracks the pre-computed dip-azimuth field that is supplied in the formof a (dip-) SteeringCube. The steering data generally determines the quality of theresulting HorizonCube.

- 161 -

The auto-tracker used to track in a dip-field works for both 2D and 3D seismicdata. Tracking in a dip field has several advantages: Firstly, the dip field is con-tinuous. Even if amplitudes vary laterally, the dip continues. Second, the dip fieldcan be smoothed before applying the tracker, which enables the controlling of thedetail that needs to be captured. The auto-tracker is applied to a target intervaland generates hundreds of horizons that are separated on average by a samplingrate. The result is called a HorizonCube. The comparison between conventionalamplitude based tracking and dip-steered tracking with SteeringCube is presen-ted in the figure above.

The following exercises are showing an application in 3D. The workflows arevery similar in 2D.

HorizonCube Types

Two types of HorizonCubes are created in OpendTect:

l Continuous HorizonCube: Contains events (or horizons) that do not ter-minate. All events are continuous throughout the entire volume. They maycome very close together (at unconformities and condensed sections) butthey can never cross each other.

l Truncated HorizonCube: Contains events that terminate against otherevents.

Both cubes have their own applications for visualization and also for model cre-ation. The advantages are also briefly explained in the following pictures.

Two types of HorizonCube based on their geometrical configuration.

- 162 -

HorizonCube Modes

A HorizonCube can be created with two different modes:

l Data driven: The data driven mode creates a HorizonCube that is guidedby the SteeringCube, in turn computed from the seismic data. Thus it will fol-low the geometries of seismic reflections. It is the preferred mode to buildaccurate sub-surface models and interpret the seismic data within a geo-logic framework.

l Model driven: The model driven mode is a way of slicing the seismic datarelative to the framework (input) horizons. There are three model driven sub-modes:

Three different model-driven modes to create a HorizonCube.

HorizonCube Tools

The following tools are available in OpendTect for performing different manip-ulations on the HorizonCube:

l Add more iterations: To fill “gaps” in the HorizonCube.l Convert to SteeringCube: Convert the HorizonCube into a dip-azimuthvolume (SteeringCube).

l Edit: Use either error-based or linear-based methods to edit events in a Hori-zonCube.

l Extract Horizons: Extract horizons from the HorizonCube (stored as hori-zon data).

l Get Continuous HorizonCube: Converts a truncated HorizonCube into acontinuous HorizonCube.

- 163 -

l Grid HorizonCube: Use various algorithms to fill unwanted holes in extrac-ted horizons.

l Merge HorizonCube:Merges multiple HorizonCubes either vertically or lat-erally. The vertical merge is useful for bigger surveys. For instance, if youhave three packages, you may run package 1 on machine 1, package 2 onmachine 2, and so forth. Then you can merge the HorizonCubesvertically toget a single output. This will speed-up the processing time when comparedto running a single HorizonCube with three packages.

l Add or Recalculate 2D Line (2D HorizonCube):Modify the HorizonCubeby adding more 2D lines or add further horizons and faults.

l Modify or Recalculate 3D Package (3D HorizonCube):Modify a Hori-zonCube by adding more horizons/faults.

l Truncate HorizonCube:Operation to remove parts of the HorizonCubebased on the event’s density (number of events within a defined time gate).

HorizonCube Inputs

The following section explains the required inputs to process a HorizonCube.Requirements include a pre-computed SteeringCube and framework horizons,whilst fault (planes or sticks) are optional.

l Pre-Computed SteeringCubeSteeringCube is a dip-azimuth volume and can be considered as the heart of theHorizonCube.

A good-quality SteeringCube will usually result in an equally good-quality Hori-zonCube. However, our experience suggests that in order to create a good Hori-zonCube, one is required to pre-compute possibly 2-3 different SteeringCubesand evaluate them by varying the HorizonCube parameters. The best Hori-zonCube is then picked by quality controlling the results. Understanding the Steer-ingCube is thus paramount towards a successful HorizonCube.

The simplest way to understand the SteeringCube is to first know the seismicdata that you are dealing with. Visualize the seismic data by scrolling theinlines/crosslines or in a volume. Focus on an interval of interest and check theareas of good and bad quality. Get an overview of whether the data quality ispoor, fair or good. If it is poor, you can expect a poor SteeringCube and thus inturn a poor HorizonCube output. Another way of looking at the SteeringCube is tolook at the geologic complexities. If the data is too complex geologically e.g. con-tains flower structures, you might not be successful.

- 164 -

In all cases, we suggest various workflows to improve the seismic data. There arethree major workflows that have been tested around the globe and are foundalways a useful step to create a SteeringCube:

1. Smooth the seismic data by applying a mild post-stack dip-steeredmedian filter (Chapter 5). Such a filter improves the quality of seismic at asub-volume scale e.g. area of 3 by 3 traces.

2. Improve the vertical resolution of the seismic by sharpening the wavelet.We normally use the Seismic Spectral Blueing (a method to enhance thevertical resolution) operation to do this. (Chapter 6).

3. Apply a band pass filter on the seismic data to discard the high frequencynoise. It is often a valuable step if you are dealing with a high frequencynoise and you want to create a HorizonCube which follows the major seis-mic events only.

Computationally, creating a SteeringCube is a slow process if dealing with a data-set of several dGB’s. Therefore, it is advisable to pre-process the SteeringCubebefore you do anything else. You can run such processing by splitting the jobs onmultiple machines.

To read more about the best settings and parameters for computing a Steer-ingCube, please go to the exercises section of this chapter.

Which SteeringCube algorithm is suitable for HorizonCube processing?

In our experience, the FFT (standard; Fast Fourier Transformation) algorithm ofdip estimation is preferred for horizon tracking or HorizonCube processing with adrawback of slowness. We recommend using the BG (phase-based) algorithm fordata conditioning and attribute analysis. This implies for both 2D as well as 3Dseismic cases.

What are the best parameters to start experimenting with various 3D Steer-ingCubes for HorizonCube?

Case 1: Assuming that the zone of has a mean frequency ranging between 25-40Hz.To create the initial detailed SteeringCube, the following parameters are good tostart with:

l Calculation [inl, xl, z] = [2,2,5]l Filtering [inl, xl, z] = No filtering if the input data is alreadysmoothed through dip- steered median filtering (DSFM).

- 165 -

You can then progressively filter this output and process the corresponding Hori-zonCubes e.g.: (using as input the SteeringCube [inl, xl, z] = [2,2,5])

l Create several filtering results [1,1,3], [1,1,5], [1,1,7],… (if thedata is not noisier).

l Or create several filtering results [2,2,3], [2,2,5], [2,2,7], … (if thedata is noisier).

Case 2: Assuming that the zone of has a mean frequency is lower e.g. 20Hz orbelow.To create the initial detailed SteeringCube, the following parameters are good tostart with:

l Calculation [inl, xl, z] = [2,2,7]l Filtering [inl, xl, z] = No filtering if the input data is alreadysmoothed through dip- steered median filtering (DSFM).

You can then progressively filter this output and process the corresponding Hori-zonCubes e.g.: (Input SteeringCube [inl, xl, z] = [2,2,7])

l Create several filtering results [1,1,3], [1,1,5], [1,1,7],… (if thedata is not noisier).

l Or create several filtering results [2,2,3], [2,2,5], [2,2,7], … (if thedata is noisier).

What are the best parameters to start experimenting with various 2D Steer-ingCubes for HorizonCube?

The settings for the 2D are much similar to 3D seismic datasets. The only dif-ference is that the calculation and filtering step-outs are Traces and Z- samples.Therefor you can use the same suggestions as answered for the previous ques-tions.

l Framework HorizonsFramework horizons (2D/3D) are the conventionally mapped horizons (3D grids /2D horizons) that serve as a geologic constraint to form a vertical boundary for aHorizonCube. Note that at least two framework horizons are needed to form apackage/sequence. The HorizonCube is always computed within given two ormore packages. So, if three framework horizons are provided, you will get a Hori-zonCube with two packages only.

- 166 -

The data-driven HorizonCube is dependent on provided framework horizons. Ituses them as a relative thickness boundary that cannot be crossed by an auto-mated HorizonCube event. Nevertheless, the automated events may be con-verged at the framework events. In some cases, such convergences couldhighlight key geologic features: pinch-outs, terminations, levees etc.

Notes and Tips:l A horizon with holes will result in a HorizonCube with holes. Thus, it is sug-gested to fill the holes by gridding horizons with undefined areas.

l Two horizons might have different geometries (boundary ranges). In suchcase the lower boundary would be used as an outer boundary of the Hori-zonCube.

l Two horizons are also used to define an automated start position (a seedposition) to track events. Tracking can in that case be started from the depos-itional centre which is the position with the thickest isopach value.

Framework horizons should be free of holes and should not cross. Optionally,they may stop at the faults. This is the Data Preparation done via the Hori-zonCube Control Center.

Does HorizonCube follow the framework horizons while tracking in a pack-age?

The framework horizons are used to calculate the starting points for various iter-ations. However, the tracked does not follow the framework horizons while track-ing. It follows the dips within the frameworks. The tracker only makesconvergence of the tracked events with the framework if the dips are making sucha case.

Can both framework horizons have different geometries?

We do not recommend using such horizons. You may end up with unexpected res-ults such has HorizonCube stopping at a bigger hole, no HorizonCube, or youmay not be able to process a HorizonCube because the start position lies in ahole. We recommend using the horizons that have common spatial geo-metries/extension, grid spacing, no holes. OpendTect has several tools to performsuch actions.

I have two horizons crossing each other. I want to use them a frameworksfor HorizonCube. Can I solve the crossings in such horizons?

- 167 -

Yes! See the data-preparation tools available in the HorizonCube control center.

Fault Planes & Sticks

Fault Planes (3D) or faultsticksets (2D) are optional inputs that can be used whencreating a HorizonCube. Faults serve as structural boundaries along which thethrow is automatically computed using the input framework horizons and a givenfault plane/stick. In OpendTect, there is an additional data preparation step tomake the framework horizons “water-tight” with the faults. There is no limitation onnumber of faults or sticks. One can still process a HorizonCube for the intervalswhere the faults are absent.

The 3D HorizonCube Creator dialog will require Faults whereas the

2D HorizonCube Creator dialog will require FaultSticksets as an input.

EXERCISE 2.4.1a: Create Horizons from SteeringCube

(Optional)Horizons can be tracked via several methods in OpendTect. Traditional amplitudetracking can be time consuming, especially if your data is not of the highest qual-ity. The new dip-steered horizon tracker uses the steering volume to auto-trackthe bounding horizons. This method is ultrafast, and can produce multiple full sur-vey 3D horizons in a matter of seconds.

Objective:Track 3D seismic horizons from a SteeringCube.

Workflow:Use inline 250, displaying 4 Dip-Steered Median Filter.

To accelerate the process of horizon extraction, you may want to pre-

load the SteeringCube into the memory.

Survey > Preload > Seismics …

In: Seismic Data Pre-load Manager, press Add Cube button

Select: 3b Steering FFT225 MF113

- 168 -

Display an inline: Tree > In-line > Mouse click > Add default data….

1. There is one large fault present that disturbs the prograding interval and it isalready interpreted.

2. Display Fault A: in the tree, right-click on Fault > Add… and select Fault Afrom the pop-up window.

3. Display it only at the section by clicking on Fault A > Display and selectingthe Only at sections option.

- 169 -

4. Launch HorizonCube control center: Processing > HorizonCube > 3D…5. In the HorizonCube 3D Control Center: Horizons from SteeringCube >

Create > Go button…

6.In the pop-up window, select the SteeringCube to use to create your horizons:3b Steering FFT225 MF113 in this exercise7.Select the Fault to be considered by the algorithm when tracking the horizon:select Fault A.8. Pick seeds: You can pick one seed per fault block. The algorithm will use eachseed as starting point for the interpretation.

a. In the Create Horizon(s) from Steering Cube window, click on the emptyHorizon 1 cell.

To rename the horizon, double click in the Horizon cell e.g. <steering

horizon 1> and type in a name.

b. Click the Pick seeds icon . This will turn your pointer in the scene into across-hair.

c. Click on only one point on the horizon you wish to track.

If you have faults in your area, you may select one seed per fault block.

- 170 -

Optimal results are achieved if you make your seed picks as far away

from the fault edges as possible, so as to let the horizons grow organ-

ically around the fault.

d. To track another horizon: click the Add new horizon icon and then pickone point on the seismic horizon you wish to track.

e. When you have made one seed pick for each horizon you wish to track,click the Pick seeds button to turn the mouse back to a pointer.

f. Click the Apply button to start auto-tracking the horizon(s) using Steer-ingCube. It will appear in the scene once it is finished.

g. Optionally, press the save icon in this window to Save these horizons.

EXERCISE 2.4.1b: Create a data-driven HorizonCube

Objective:Create a data-driven HorizonCube in a prograding interval to understand itsdepositional history using the principles of sequence stratigraphy. It will be a con-tinuous HorizonCube type.

Workflow:

To save processing time - especially in the next exercise (adding iter-

ations) - you can limit the processing range to inline 400-450.

Step 1 – HorizonCube Processing

1. Follow Processing > HorizonCube > 3D… and click on the Create buttonfor the New HorizonCube

- 171 -

2. In the HorizonCube Creator 3D, click Read horizons button and select atleast two horizons to be your bounding surfaces. See below:

- 172 -

This will populate the HorizonCube calculation mode.

Trimmed horizons

These are the horizons that are water-tight with the faults. It first

removes a part of the horizon with a given radius and creates a gap.

Then it grids the gap either using the SteeringCube or a conventional

gridding algorithm.

The fault throw is automatically calculated by the tracker based on the

provided horizons and faults. This process is considered as a data pre-

paration step and is done via the HorizonCube control center > Data

preparation > Trim horizons against faults.

The illustration of a trimmed vs. un-trimmed horizon is shown below.

3. In this exercise, we want to create a data driven HorizonCube that issteered by a SteeringCube.

- 173 -

In the first package:

a. Change the mode to Data Driven.b. Press the Settings buttonc. In the settings dialog:

l Select 3b Steering FFT225 MF113 as input Steer-ingCube

l The Start at drop-down menu has several options.Each of these allows you to define where the initialstarting points for each horizon will be located. Thiscan be a very influential step, so be sure to try vari-ations of this option if you are not satisfied with yourresults.

l In this exercise: Set the start at position to beCentre.

l The Advanced options are used to edit more settings.There you can decide to create a Continuous or Trun-cated HorizonCube or change the step. In this exer-cise: Leave them as default.

- 174 -

d. Click OK to the settings dialog.

In case of multiple packages you will notice that the settings you made

in the first package have also been applied to the second package.

You can change the settings for the second package. Unselect the

Apply to all packages option before clicking on Ok when defining the

settings of the first package.

4. Select Fault A in the HorizonCube Creator (If there are more faults, you mayselect multiple faults as well).

5. Give a name to your HorizonCube, e.g. HorizonCube FFT 225 MF 113.

6. Click on the Analyze button. This will quickly test all of the settings you havemade to see if they would result in a successful HorizonCube. The reportshould end with: All packages passed successfully otherwise there is some-thing wrong with your parameters. The brief description of the analysis inthe report can help you to locate the problem.

7. If the packages are passed successfully, dismiss this report, and click theRun button on HorizonCube Creator window. A separate window will openup in where you can follow the progress of the HorizonCube processing.You can then dismiss the HorizonCube Creator window.

You can continue to work in OpendTect while this is calculating.

8. Once the processing is complete, i.e. when the Finished Batch Processingmessage appears in the processing window, close this window using the

icon.

Step 2 - Displaying the HorizonCube

9. Display Inline 425 in the scene with 4 Dip Steered Median Filter

10. Rick click on the inline number and choose > Add > HorizonCube display

- 175 -

11. Select the HorizonCube FFT 225 MF 113 that you just created.

An overlay of a data-driven HorizonCube.

12. Once the HorizonCube is displayed, you can change the display properties:in the tree, right click on the HorizonCube name > Display > Properties.

- 176 -

Display properties of a HorizonCube.

HorizonCube SliderThe HorizonCube slider is a very useful tool to investigate your data and to makedetailed observations of the depositional history of your sedimentary basin.

13. At this point, slide the HorizonCube events up/down to QC the results using

the HorizonCube slider .

- 177 -

Using the HorizonCube slider, you can scroll the events up/down withan interactive update in the scene.

Step 3 - Fill gaps by adding iterations

In the previous step, you might have noticed several gaps in the HorizonCube res-ults. These gaps are filled by adding more iteration(s) to an existing HorizonCube.The gaps are filled by finding the gaps based on a given advanced setting (seebelow). If the gap size is bigger than a given thickness and number of traces, itwill start filling them by inserting a new start position at a position of maximumthickness.

HorizonCube advanced options: default settings for finding gaps arehighlighted in red coloured rectangle. These are set while processingthe HorizonCube. You cannot change them once you have created aHorizonCube. We mostly recommend using the defaults and addingthe subsequent iterations later on.

14. Follow Processing > HorizonCube > 3D… to access the HorizonCube 3DControl Center.

15. In the drop-down Tool menu, select Add more iterations and click on Go but-ton.

- 178 -

16. In the pop-up window:l As input HorizonCube, select the HorizonCube you created in the pre-vious step.

The first column tells you how much iteration is already processed in thisHorizonCube.The second column allows you to change the number of additional iter-ations you will add in this step.

l You can add a different number of iterations to the different packagesor just add for one. For this exercise, leave this number to 1 for eachpackage.

l Save with the same name as before, but add a notation that lets youknow how many iterations it will have after this step.

- 179 -

Note that the gaps are filled by adding more data-driven horizons during the 2nditeration. Further gaps can be filled by adding one or more iterations.

Step 4 - Change the HorizonCube displayed

You can change the display of a HorizonCube in two manners:

l In the tree:

- 180 -

l In the HorizonCube Control Center:

EXERCISE 2.4.1c: 3D Visualization of a HorizonCube

To visualize a HorizonCube in 3D, you will have to use the 3D slider. It is an add-on to HorizonCube to perform an analysis in 3D along the events:

l Make interactive thickness maps,l Extract geo-bodies,l Extract horizons,l Perform stratal slicing using seismic attributes.

Objective:You will study how a prograding system behaves in 3D by making a set of thick-ness maps or geo-bodies.

General workflow:

A generalized workflow is given to study depositional shifts using a series of thick-ness maps.

1. Identify the packages that you want to study using a continuous Hori-zonCube. This can be done on vertical sections e.g. inlines or random lines.You may want to note down the event ranges defining the packages.

2. Create thickness maps using the 3D slider and produce screenshots or 3DPDFs.

3. Alternatively, prepare geo-bodies outlining the thicker regions as depo-centers or preserved areas.

- 181 -

Workflow:

In this exercise, we will be using a pre-processed continuous HorizonCube thatcovers the entire survey area.

1. Display inline 425 with an overlay of HorizonCube:a. in the tree, right-click on Inline > Add default data.b. Right click on Inline 425 > Add > HorizonCube display and choose

Continuous_HC_Full.2. Launch a 3D slider: Processing > HorizonCube > HorizonCube 3D Control

Center… and press the button 3D Slider.3. In 3D slider:

a. Pre-load the HorizonCube: on Pre-loaded HorizonCube > Press i andselect Continuous_HC_Full. Now it will preload this HorizonCube intothe memory and will display two blank horizons in the scene.

b. Display data > Top-Base isopach…..to display colour coded thick-nesses between the top and base in TWT units.

c. Set check Link to 2D displays….to link the slider with the scene suchthat the slider positions are reflected on inlines, crosslines, and ran-dom lines.

d. For the Top slider, set the Display > In full to None.

- 182 -

An example illustration of 3D slider settings for the continuousHorizonCube.

4. Make thickness maps:a. Type in for Top slider = 302 (press ENTER to update) and Base slider

to 310 (press ENTER to update).b. Press the Calculate button. The resultant thickness map will appear in

the scene on the base horizon.c. Type in for Top slider = 285 and Base slider to 302.d. Press the Calculate button.e. Type in for Top slider = 275 and Base slider to 285.f. Press the Calculate button.g. Type in for Top slider = 265 and Base slider to 275.h. Press the Calculate button.i. You will observe a net westward shift of the thicker regions.

- 183 -

An example illustration of a thickness map with transparency.

Several geo-bodies are extracted using 3D Slider.

Make geo-bodies maps:a. With current slider positions i.e. top = 265 and base = 275, you may want to

outline the thicker regions into a geo-bodies for further analysis.b. We will apply a transparency threshold for all thinner regions and will try to

preserve an outline of thicker region.

i. At the bottom of the 3D slider, set the transparency. Threshold = 0.02s(Enter) and change the transparency disabled to Transparent belowthreshold.

ii. In the histogram, a red line will appear. You can interactively move itto adjust the display.

iii. Press Body button.

l Automatic fillingl Threshold = unchanged

- 184 -

l Body value = Above threshold, suggesting to create the body outliningthe values above the threshold.

iv. This will create a geo-body and will add it in the tree / scene

c. If you like the body, you can save it via the tree > Body > <New MC 1> Rightclick > Save as….

You can repeat this exercise from # 5 with different top/base values to createmore geo-bodies.

EXERCISE 2.4.1d: Truncate a HorizonCube

The HorizonCube has two output types: continuous and truncated. These outputshave very different applications. Typical uses for a continuous HorizonCubewould be 3D model building, and a Truncated HorizonCube would be used in aWheeler Scene for viewing depositional stacking patterns. Truncation is based ona user-specified density of events per sample interval. A HorizonCube originallymade as continuous can be truncated, and vice versa. We will practice creating aTruncated HorizonCube from a Continuous HorizonCube.

Objective:Truncate a HorizonCube to prepare Wheeler diagrams and perform sequencestratigraphy.

Workflow:1. Processing > HorizonCube > 3D > HorizonCube 3D Control Center and

select Truncate HorizonCube in the Tool menu.2. Select the original (continuous) HorizonCube as input cube.3. Leave the area sub-selection and minimum spacing as default.4. Name your output cube with a similar name as the original but add a nota-

tion to remind you that this one has been truncated, e.g. HorizonCube Trun-cated.

- 185 -

5. Click Proceed to process the HorizonCube.

EXERCISE 2.4.1e: Extract Multiple Horizons from a HorizonCube

The ultimate goal of the HorizonCube is to have horizons essentially everywhere.These can be used as input for many other advanced features. You can save anyhorizon from the HorizonCube as a stand-alone item.

Objective:Extract horizons from a HorizonCube.

Workflow:1. Inline > Add default data and right click on 425 > Add HorizonCube overlay.

The active HorizonCube is automatically displayed.2. Processing > HorizonCube > 3D > HorizonCube 3D Control Center and

select Extract Horizons in the Tool menu.

- 186 -

3. Scroll up and down with the slider, watching the HorizonCube in the scene.

4. When you locate a horizon you want to extract: click on the Pick Horizon but-ton.

5. A second window will appear: give a name to the new horizon and click Ok.

6. Once the horizon picked and named, its location is marked with (in thiscase) a grey line in the Extract Horizon window.

7. Optionally, repeat this process for multiple horizons,8. Click Proceed

Your horizon(s) is(are) now stored and ready to load.

2.4.2   Sequence Stratigraphy Interpretation System (SSIS)

What you should know about SSIS:l SSIS is a commercial plugin by dGB that operates on a HorizonCube.l SSIS supports Wheeler transformations and Systems Tracts Interpretation.

- 187 -

Details:In essence, sequence stratigraphy is used to provide a chronostratigraphic frame-work for correlation and mapping and for stratigraphic prediction (Emery andMyers, 1996). Although sequence stratigraphy has proven to be a powerful instru-ment, and despite major advances in concepts since its introduction in the nine-teen-seventies, sequence stratigraphy has not lived up to its potential because ofthe lack of supporting software tools. OpendTect’s SSIS plugin came to the mar-ket with the aim of filling this gap.

Wheeler diagrams and wheeler transforms can be powerful tools to aid insequence stratigraphic interpretations. Non-depositional or erosional hiatuses arevisible, the lateral extent of stratigraphic units can be determined at a glance, anda clear understanding of the lateral shift in deposition over time can be estab-lished. The Wheeler transform is constructed, by flattening each horizon, thusenabling the user to study seismic data, and its derivatives (attributes or neuralnetwork outputs) in the Wheeler domain in three dimensions. Previously, Wheelerdiagrams were constructed by hand, making this a time consuming process. Thisis unfortunate because the Wheeler diagram, or Wheeler transformas its seismic counterpart is called, is a very valuable tool to gain insight and toextract additional information.

The Sequence Stratigraphic Interpretation System (SSIS) plug-in to OpendTectallows interpreters to automatically create a Wheeler transform in which they canstudy the depositional history of the area through flattened horizons, showing thestacking patterns including depositional hiatuses and condensed sections. Usingthis added feature, interpreters can make more informed decisions about seismicfacies and lithofacies predictions, thus helping to identifying potential

- 188 -

stratigraphic traps.

SSIS will only be of use if you have already calculated your HorizonCube. If youcreated a continuous HorizonCube, you will need to truncate this to see depos-itional variations in the Wheeler scene. Both creating a HorizonCube and trun-cating an existing one are covered in the previous section of this training manual.

Lap-out patterns & stratal terminations

While it is not a requirement as part of the workflow to perform this step each time,but it is considered as a good practice. Annotating the stratal terminations in yourdata before making your interpretations can lead the observations towards properinterpretation.

Annotations are graphical interpretation tools that are available in OpendTect dur-ing the whole workflow. They can be a great help at the start of an interpretationwhen tracking bounding surfaces, or when making an initial interpretation.

The annotations comprise of three basic tools: Arrows, images and scale bar. Thearrows are intended to indicate lap-out patterns or stratal terminations, but can beused to highlight any feature. Seismic data can be animated with pictures to makecommunication easier and more direct with colleagues who are working on thesame project. The scale bar allows you to very easily add scale information.

The types of stratal terminations are truncation, toplap, onlap, downlap, andofflap. They provide diagnostic features for the recognition of the various surfacesand systems tracts. “Stratal terminations also allow inferring the type of shorelineshifts, and implicitly the base level changes at the shoreline. For example, coastalonlap indicates transgression, offlap is diagnostic for forced regressions, anddownlap may form in relation to normal or forced regressions.” (Catuneanu,2002):

- 189 -

Types of stratal terminations

Stratal Termination Shoreline shift Base levelTruncation Fluvial FR FallTruncation Marine FR, T Fall, RiseToplap R StandstillApparent toplap NR, FR Rise, FallOfflap FR FallOnlap, fluvial NR, T RiseOnlap, coastal T RiseOnlap, marine T RiseDownlap NR, FR Fall, Rise

- 190 -

Selecting unconformities

EXERCISE 2.4.2a: Annotate Stratal Termination (Optional)

Objective:Annotate stratal terminations and lap-out patterns.

Workflow:

1. Load inline 425 and click on the '+' next to Annotations to expand the selec-tion.

2. Click on Arrows and select New Arrows Group…3. Name the arrow group to Stratal Termination

- 191 -

4. Click on the seismic data where you see a stratal termination or lap-out pat-tern.

The first click adds the arrow-head; the second click adds the direction

For greater control over the direction move away from the arrowhead

before making the second click

Terminations are better visible if you use one-sided arrowheads (Prop-

erties option in the tree menu)

Change the type, color, width and size of the arrow via right-click on

your New Arrow group > Properties...

An example of a downlapping arrow-head.

5. Make a complete interpretation of inline 375 by indicating all stratal ter-minations, highlighting features with text boxes, etc.

EXERCISE 2.4.2b: Evaluate Stacking Patterns

Objective:Evaluate stacking patterns using the HorizonCube Slider.

Workflow:1. Overlay inline 425 with a HorizonCube by right clicking on 425 in the tree >

Add > HorizonCube display.

- 192 -

2. Click on the HorizonCube Slider icon , and a new window will openwith a slider for the top and the bottom directions.

3. Play with the sliders, which will interactively adjust the display of the hori-zons, to understand the depositional history of the system as reflected bythe auto-tracked horizons.

4. Display this on a crossline at the same time and consider the stacking pat-terns.

Alternatively, you can display a random line which is oriented in a

depositional dip direction – and overlay it with the HorizonCube.

Wheeler Transformations

The Wheeler transformation is flattening of the geometries in a HorizonCube andsorting them chronological such that the y-axis represents a continuous seriesreflecting the number of events in a HorizonCube. Using this as a tool in makingsystem tracts interpretations puts the interpreter at an advantage since the stack-ing patterns, hiatuses, and erosional features become so apparent.

Data is best studied simultaneously in the Wheeler domain and in the normal ordepositional domain. In the depositional domain, structural features are visible butother features stay hidden. Several of these features are exposed in the Wheelerdomain, but this domain lacks the structural aspect. One of the most apparent fea-tures in the Wheeler transform is that hiatuses are visible. Both non- depositionalevents and erosional truncations can be distinguished (see figure c below).

Stratigraphic thinning or condensed sections can also be identified in theWheeler transform. During the deposition in areas of condensed sections, sed-imentation rates are very low causing stratigraphic events to merge below seismicresolution so that they cannot be auto-tracked. Therefore, even though strati-graphic thinning might not be a true hiatus, they do show up in the Wheeler trans-form (and the original Wheeler diagram) as such (see figure c below).

Additionally, the lateral extent of stratigraphic units or individual chro-nostratigraphic events can be determined with ease in the Wheeler transform.This can be a daunting task in the depositional domain, especially when no Hori-zonCube is available. The Wheeler domain is thus ideal for the study of the devel-opment of deposition over time, helping to answer such questions as “how doesdeposition shift (laterally) over time?”, “what is the lateral and temporal distributionof the packages?”

- 193 -

Depositional domain a) HorizonCube b) Systems tracts interpretation. Wheelertransforms: c) HorizonCube d) Semi-transparent systems tracts overlaying theseismic data e) Seismic data f) Systems tract interpretation.

EXERCISE 2.4.2c: Make a Wheeler Display

Objective:Wheeler transform (flatten) seismic data and co-render the flattened seismic withthe flattened horizons of a truncated HorizonCube.

Workflow:

- 194 -

1. Follow Scenes > New [Wheeler]2. A new scene with a new tree opens up.

a. Load the same inline (425) into the new Wheeler scene with a defaultdata: Tree > Inline > Add default data …

b. Overlay this inline with the continuous HorizonCube: right-click on 425> Add HorizonCube display… and select Continuous_HorizonCube_Full.

The selection of a HorizonCube in one scene applies that Hori-

zonCube to the other scene as well. This applies in both directions

(i.e.: updating the HC in the Wheeler Scene will change the HC in

Scene 1, too.)3. Viewing a continuous HorizonCube in the Wheeler scene is not very inter-

esting so the first thing to do is to truncate the HorizonCube display.a. Right-click in the tree on the HorizonCube entry and select Display>

Properties.b. Toggle Truncate HorizonCube on and test truncating with different

densities (e.g. 2, 10, 20) and chose the one you like best.4. Maximize the Wheeler scene so that it is all that you see in your window.

Use the HorizonCube slider to scroll up and down through your data in theWheeler display. Notice the areas of non-deposition and erosion will bestretched in the Wheeler scene because this is a continuous HorizonCube.However, with the truncated HorizonCube display, you can easily recognizethem. This is the reason why, we recommend using the truncated Hori-zonCube for Wheeler transformation and sequence stratigraphy.

Do you see the prograding system?

Now you will view the HorizonCube in the Wheeler scene and the normal scenesimultaneously: go to Scenes> Tile > Vertical.This will stack one scene on top of the other, and you can use the HorizonCubeslider again, this time viewing the two scenes simultaneously.

Additional Note:In the Wheeler scene, flattened seismic data can be displayed by adding ele-ments (inline, crossline, Z slice, random line or wells) in the tree. Such trans-formation is done on the fly. All attributes and neural network outputs arecalculated in the normal domain first and then transformed to the Wheelerdomain. Because of this transformation, quickly scanning through your Wheeler isonly possible after creating a stored Wheeler Volume (Processing > SSIS >Create Wheeler Output 3D).

- 195 -

Systems Tracts Interpretations

What you should know about systems tracts interpretations:

l Definitions of terms such as high-stand, low-stand, sequence boundary etc.depend on the sequence stratigraphic interpretation model that is beingused.

l OpendTect supports all industry-standard models.l The software has built-in intelligence that depends on the model used. Forexample sequence boundaries are marked, boundary names are proposedand a relative sea level curve is reconstructed based on interpreted systemstracts.

l You can create your own model to divide an interval of interest into pack-ages. If you do this you obviously loose the built-in intelligence of standardmodels.

Choice of System Tracts Model

Within the sequence stratigraphic community several different sequence modelsare currently used, each with its own set of terminologies for systems tracts andstratigraphic surfaces and with their own position of the sequence boundary(Catuneanu, 2002). The software is not bound to any one of these models, sincesystems tracts terminology and the position of a sequence boundary are user-defined variables.

A systems tracts interpretation is based on user defined geo-time intervals. A sys-tems tract is thus bounded by two chronostratigraphic events selected by the user.All intermediate chronostratigraphic events are assigned to the interpreted sys-tems tract. Similar to the HorizonCube, an overlay of interpreted systems tractscan be made on inlines and crosslines.

This flexibility also allows sequences to be sub-divided into user defined depos-itional packages, with an individual color and name for each package, when sys-tems tract interpretation is impossible or difficult.

Default System Tracts Model

As standard, we subdivide a full sequence into four systems tracts: Falling stagesystems tract (FSST), Lowstand systems tract (LST), Transgressive systems tract

- 196 -

(TST) and Highstand systems tracts (HST) (Figure below). The interpretation isbased on the following basic principles:

l A transgression is a landward shift of facies and shoreline, whilea regression is a seaward shift of facies and shoreline. (Catun-eanu, 2002).

l A transgression or transgressive systems tract is characterizedby a retrogradation and aggradation. This occurs when base-level is rising and more accommodation space is created than isconsumed by sedimentation.

l Regressions can be subdivided into normal and forced regres-sion:

o During forced regression base-level is dropping, forcingthe system to prograde. Forced regression is characterizedby progradation and incision (erosion).

o During normal regression base-level is rising but the con-sumption of accommodation space by sedimentationexceeds the creation of accommodation space by thebase-level rise. Normal regression occurs during the firstand last stages of base-level rise and is characterized byprogradation and aggradation. The lowstand systems tractand highstand systems tracts are both normal regressiondeposits.

- 197 -

Using the tiled view of both the normal and Wheeler scenes, an interpreter canquickly produce a rough interpretation based on viewing one line, and then viewthe interpretation in other areas to QC.

- 198 -

EXERCISE 2.4.2d: Systems Tracts Interpretation

Objective:Perform a systems tracts interpretation with the SSIS plugin.

Workflow:

1. Using the arrows created on inline 425 in the annotations exercise you willmake a first pass SSIS Interpretation. Toggle on your arrows group, foundunder the annotations section in the tree.

2. Open the SSIS interpretation window by clicking on this icon:3. A blank interpretation window will open, with the familiar HorizonCube

Sliders on the left side, and interpretation column, base level curve, andtimeline in the white pane on the right.

4. First, open the Sequence Models selection window to view your options, by

clicking the tools icon in the SSIS Interpretation window.l View the options of the sequence models available. For this exercise,the default model (Depositional Sequence IV) will be used.

l Close out this window when you have finished viewing the options.

The sequence models are setup in OpendTect according to the hier-

archical description of Catuneanu, 2002.

- 199 -

5. To begin your interpretation, you will first use the HorizonCube sliders:l Slide the top slider all the way to the bottom while watching theWheeler and normal scenes.

l Slowly drag it up until you find a breaking point that would indicate adifferent system tract.

Perhaps you have already marked this with one of the arrows?

6. Once you have located a position where you want to insert a boundaryusing the top slider: click the Insert button located above the top slider. Anewly inserted <Unassigned> boundary has been inserted in the systemstracts column.

7. You may now assign a system tract to this area by right clicking in the areabelow the newly inserted boundary.

If you are not sure what system tract you would like to assign, you may

skip this step or assign as undefined at this time, and come back later

and make assignments (The stratigraphic surface will be assigned

according to the selected sequence model).8. Following the same procedure, identify other systems tracts and interpret

the entire package.9. Add the systems tract display to inline 425 in the normal domain: right-click

on 425 in the tree > add Systems tracts display.10. When you are done, press the Ok/Save button to save your interpretation.

- 200 -

EXERCISE 2.4.2e: Statistical (Thickness) Curves

The base-level curve in OpendTect is a theoretical curve which is sketched onthe basis of the selected sequence model. You will notice that there are somesequence models for which there is not base level plotted. Hence, a base-levelcurve is not plotted in the SSIS display for when those sequence models aremade active.

Statistical curves are an add-on to express data-driven nature of base-level vari-ations reflected on thickness variations. These are thickness curves plotted nextto the base-level curve. They also help in explaining the preserved sedimentationin conjunction to the theoretical base-level variations. We allow options to cal-culate such curves in a user-defined region. For instance, you can make two setof curves in a simpler case: one in the coastal regions and the other one in thebasin / depocenters. When these curves are plotted on a 2D view with a systemstract column, you can easily explain nature and a qualitative estimate of sed-imentation to explain the depositional history.

Objective:Save identified system tracts boundaries as normal horizons in OpendTect.

Workflow:1. You should complete the SSIS interpretation and make a stratigraphic

column such that you have assigned the systems tracts to the entire

- 201 -

HorizonCube. (see previous exercise)2. In the SSIS interpretation window, press the icon called statistics3. In the pop-up window, choose an area within which you would like to estim-

ate the thickness and proceed. Once the process is finished, the thicknesscurve will be plotted.

4. Optionally, change the statistics of the display by right-clicking on the curve.

Change statistics of the curve.

EXERCISE 2.4.2f: Extract Stratigraphic Surfaces

You can save the SSIS surfaces as OpendTect horizons.

Objective:Save identified system tracts boundaries as normal horizons in OpendTect.

Workflow (Option 1):1. Choose one of your newly interpreted horizons, and right click on the

surface name, select Save as Surface.2. You may provide a new name for your surface in the Output Surface

field, and select Ok.3. Go to the tree in the normal scene, right click on Horizons and load

the horizon you just saved. Depending upon the speed of your

- 202 -

computer, you can select only a sub-volume to process, limiting therange between inlines 375-425.

Workflow (Option 2):1. In SSIS window, click on Save all Surface to save multiple surfaces in

one go.2. In the pop-up dialog, you will have all surfaces listed in the left list

box: you will need to add them to the right and given an output name.The added horizons will be copied to general OpendTect horizons.

EXERCISE 2.4.2g: Stratigraphic Attributes

Once you have a HorizonCube and performed interpretation on it, you can extracttwo sets of stratigraphic attributes:

1. HorizonCube Attributesl HorizonCube Data: It converts the stored HorizonCube data (anyattribute), which is natively a set of grids, into a volume.

l HorizonCube Density: It counts number of horizons within a timegate. Higher values represent denser regions such as unconformities,pinchouts, condensed sections etc.

l HorizonCube Dip: It is a dip volume derived from auto-tracked hori-zons.

l HorizonCube Layer: You can break the HorizonCube into a set ofarbitrary layers to produce an un-interpreted HorizonCube volume.

l HorizonCube Thickness: You may also find gaps in a HorizonCubeto understand their geological reasons. Often it is helpful to

- 203 -

understand complexities in a depositional system e.g. channel base,levees, mounds that are not filled by the defined iterations.

2. SSIS Attributesl Systems Tracts: It is an attribute that is extracted from interpreted sys-tems tracts. You can output:

o Common IDs – If a system tract is repeated, it will have thesame number in the output volume.

o Unique IDs – It is a count of number of systems tracts in an inter-preted interval. It counts from top to down.

o Thicknesses – It is a thickness volume per systems tract.

Objective:Define and understand the HorizonCube/SSIS attributes.

Workflow:

Launch the attribute set window .Define HorizonCube attributes with default parameters and add them in the list.a. HorizonCube density:

b. HC_Layers:

- 204 -

c. Systems Tract (Common IDs, unique IDs, Isochron):

The attribute set window should look like this figure.

3. Once you have defined these five attributes, close the attribute set windowand give a name to this set e.g. Stratigraphic Attributes.

4. Optionally, display these attributes on any inline:a. For instance, have a look at the HorizonCube density attribute and

understand its result.b. Display the Systems tracts Isochron to see thickness variations on sec-

tions.5. Or if you are satisfied, process the attribute as a volume: Processing >

Create Seismic Output > Single Attribute > 3D….

- 205 -

EXERCISE 2.4.2h: Stratal Slicing

Stratal Slicing (or proportional slicing) is a technique that constructs the inter-mediate horizons at fixed intervals between two horizons. It is an excellent tool toquickly analyze 3D seismic data by slicing through all available data.

Objective:This exercise requires at least two horizons and/or faults already available in thesurvey.1. Create a model based HorizonCube.2. Define a seismic attribute if not already processed. We normally recom-

mend using Similarity, curvature and amplitude based attributes.3. Create a Wheeler output (Optional for smaller surveys and mandatory for

bigger size surveys).4. Display the Wheeler output in the Wheeler scene using the same Hori-

zonCube.5. Slice the volume upward and downward for the z-axis to see geo-

morphologies.

Workflow:

1. Create a model driven HorizonCube (no dip steering is required).a. Processing > HorizonCube > 3Db. In the pop-up window, New HorizonCube > Create button.c. In HorizonCube Creator, Read horizons (and choose Demo 4, Demo

6 and Demo 7).d. The table will fill automatically with two packages. You define the

mode to proportional for both packages. e. Write an output name to theHorizonCube Proportional_HorizonCube and proceed.

2. Select the Proportional_HorizonCube as an active HorizonCube: Hori-zonCube Control Center > Active HorizonCube > Select.

3. Define similarity attribute in the attribute set window.4. Create wheeler volume: Processing > SSIS > Create Wheeler Output >3D.

Follow this figure:

- 206 -

Create an attribute volume for Wheeler Scene visualization.

5. Make a Wheeler scene: Scene > New [Wheeler].6. Load an empty volume in Wheeler scene: Wheeler Tree > Volume > Add.7. To view a data in this volume, right click on the volume > Add > Wheeler

data > in the next list, you will find the data that you just processed i.e.Wheeler: Similarity_from_Proportional_HorizonCube.

Stratal slicing in the Wheeler domain

References

Keskes, N. 2002. GEOTIME TM: A New Tool for Seismic Stratigraphy Analysis.VailFest; Sequence Stratigraphic Symposium; A Tribute to Peter R. Vail. 7-9March 2002.

Lomask, J., 2003. Flattening 3D seismic cubes without picking. SEG Expandedabstracts 22, 2004.

- 207 -

Overeem, I., Weltje, G. J., Bishop-Kay, C. and Kroonenberg, S. B., 2001, The LateCenozoic Eridanos delta system in the Southern North Sea Basin: a climate sig-nal in sediment supply? Basin Research 13, 293-312.

Posamentier, H. and Allen, G., 1999. Siliciclastic sequence stratigraphy – con-cepts and applications by. SEPM publication Concepts in Sedimentology andPaleontology No. 7. Tulsa, Oklahoma, Nov. 1999. ISBN: 1-56576-070-0.

Sørensen, J.C., Gregersen, U., Breiner, M.& Michelsen, O. (1997) High frequencysequence stratigraphy of upper Cenozoic deposits. Mar. Petrol. Geol., 14, 99-123.

Stark, T.J., 2004. Relative geologic time (age) volumes – Relating every seismicsample to a geologically reasonable horizon. The leading Edge, Sep. 2004,Vol.23, No. 9.

Tingdahl, K., de Groot, P. and Heggland, R. (Statoil), 2001. Semi-automatedobject detection in 3D seismic data. Offshore, August 2001

Catuneanu O., 2002. Sequence Stratigraphy of Clastic Systems: Concepts, mer-its, and Pitfalls. Geological Society of Africal Presidential Review No.1. Journal ofAfrican Earth Sciences 35.

2.4.3   Well Correlation PanelWhat you should know about Well Correlation Panel:

l The Well Correlation Panel is a commercial plugin by dGB.l It is used to pick and QC well log markers.l It can be used to create conventional well correlation panels without seismicbackdrop.

l Typically, however, a random line is created through the wells and the seis-mic is used as a backdrop to guide the interpretation.

l A HorizonCube, either 3D, or a dedicated 2D version created along the ran-dom track, can optionally be added to help correlate markers from well towell.

- 208 -

Details:A part of sequence stratigraphic interpretation (next chapter) is to integrate theseismic information with the well data. This is done in the Well Correlation Panel(WCP). The panel is an important tool for creating consistent geologic frame-works. It integrates stratigraphy, well logs, markers, mapped regional horizons,seismic and horizons from the HorizonCube in one view. It enables the user toarrive at interpretations that are consistent between the different scales of(regional) geological concepts, seismic data and well logs. Its primaryfunctionality is to pick (and/or QC) well log markers that are defined within the(regional) geological or stratigraphic framework in a consistent manner using seis-mic correlations to guide the picking. Typically, the user starts with a random seis-mic transect connecting the wells in a 3D volume. A well correlation panel isconstructed along this random track and the Well Correlation Panel is launched.

However, if the user wants to use a HorizonCube to guide the correlations it canbe beneficial to convert the random line into a 2D seismic section and to continuewith 2D mapped horizons and 2D HorizonCube. In that case 3D regionalhorizons are converted to 2D horizons (tree option under 2D Horizon) and a Hori-zonCube is created along the 2D section. When this is done, the Well CorrelationPanel is launched. Here the user picks and QC’s markers. To use all supportedfunctionality the user should build a stratigraphic framework that links (regional)well markers to seismic horizons. Both time and depth domain are supported inthe WCP module. OpendTect’s synthetic-to-seismic matching module is fully

- 209 -

integrated and is used to align all wells optimally before picking/editing markers.WCP supports various display modes including but not limited to: wells only;wells plus seismic; equidistant; connecting markers; filling stratigraphy. Unique isthe capability to display the dense set of horizons from the HorizonCube and useof the HorizonCube slider to guide detailed correlations.

EXERCISE 2.4.3

Objective:Create a well correlation panel and QC the well markers using the seismic dataand a HorizonCube to guide you.

Workflow:

1. Launch the WCP window: Analysis > Well Correlation Panel, or via theicon.

2. Select data type: 2D line3. Input data should be seismic and the line name should beWell Correlation-

Line.4. At the bottom of the window, select all wells available in the list.5. The next step is to display the WCP window... press the Go button.

Once you have launched the WCP with the seismic data and the well data dis-played as default, the next step is to make a display that you can easily use forinterpretation.6. Change the seismic color spectrum to Grey scale.

l Click on VD and from the top bar spin the color to Grey scale.7. Overlay the seismic data with the HorizonCube. Use the Tree item called

HorizonCube to display an existing HorizonCube on top of the seismic data.l Right-click on it, and Add HorizonCube display.

8. Now display a Gamma Ray (GR) log on all wells. Use the well properties

icon from the WCP window.l In the Log 1 tab, you may select Gamma Ray log with a SandShalecolorbar.

l In the Log 2 tab, you do not select any other log (Select None).l Press Apply to all wells button to display the Gamma Ray log on allwells already displayed in the panel.

9. Display the wells panel display on top of the seismic i.e. press this Settings

icon that is available at the bottom of the panel. In the pop-up dialog,

- 210 -

set check to On top option.10. Now start interpreting the depositional trends and possible systems tracts

boundaries by moving the HorizonCube slider up and down and

adding new markers .l The markers are added using the + sign icon. It will launch a pop- updialog where you will give a name.

l Once the marker is added in the list, it will become an active marker.l You will have to click inside the well area to add a marker.l To delete a marker, use CTRL + mouse click on the marker.l If you have finished interpreting markers, please press the save buttonin the markers dialog and close it.

The F03-4 and F03-2 wells are mainly targeting the coastal plain to

shallow marine type of depositional settings within the shallow interval

i.e. between 500- 1000ms. However, the wells F02-1 and F06-1 are tar-

geting the slope to deep marine settings for the same interval.

2.5 Seismic PredictionsThis chapter deals with Quantitative  Interpretation possibilities in  OpendTectusing commercial plug-ins developed by dGB, ARK CLS & EarthWorks. Theseplug-ins cover a wide range of seismic inversion and forward modeling methods.

In this chapter you will learn how to:

l Perform relative (band-limited) acoustic impedance inversion with SeismicColoured Inversion (SCI) plug-in.

l Perform  model-driven  absolute  acoustic  impedance  inversion  withDeterministic Inversion (DI) plug-in.

l Perform stochastic inversion with Multi-Point Stochastic Inversion (MPSI)plug-in.

l Predict porosity from inverted acoustic impedance and porosity well logsusing Neural Networks plug-in.

The inversion plug-ins (SCI, DI, MPSI) can be  used to invert to (Extended) ElasticImpedance volumes using the same work flows described in this Chapter. Formore extensive training in inversion, please contact dGB at info@dgbes.com.

Training of SynthRock, dGB’s plug-in for simulation of pseudo-wells and HitCubeinversion (matching stochastic pseudo-well synthetics against measured seismic

- 211 -

responses) is not included in this manual. A separate training manual exists. Formore information, please contact dGB at info@dgbes.com.

2.5.1    Relative Impedance Inversion (SCI)What you should know about Seismic Coloured Inversion (SCI):

l SCI is a plug-in by ARK CLS.l It enables rapid band-limited inversion to Acoustic or (Extended) ElasticImpedance.

l The SCI operator matches the seismic amplitude spectrum to the well logspectrum.

l Default trends can be used in the absence of well logs.l The workflow is very similar to Seismic Spectral Blueing (see section 2.3.4).

The workflow is as follows: an operator is designed for SCI using the seismic andwell data. Once the operator has been derived, it is converted to the time domainand simply applied to the seismic volume using a convolution algorithm.Our aim is to design an operator at the zone of interest (target). It is therefore desir-able to time gate the selected traces prior to generating well log spectra. Ideallyyou should use a good interpreted horizon in the target zone to guide the welldata (log traces). In this manner, the various gated log traces should have samplevalues over a similar geology. However, in our case we will just use a windowinterval instead.

Here is the workflow on how to create and apply these techniques in OpendTect:

1. Seismic: Amplitude-Frequency plot2. Smoothing of seismic mean3. Well: Amplitude-Frequency plot4. Global trend of well plot5. Design operator6. Apply Operator7. Quality Check

EXERCISE 2.5.1

Objective:Invert the seismic data to relative acoustic impedance using SCI plugin.

Workflow:

Step 1 - Launch Seismic Coloured Inversion

- 212 -

1. Open the Attribute Set window by clicking on the .2. Select Coloured Inversion in the Attribute type list.3. Define the attribute:

a. Select 1-Original Seismic as input data.b. Click on Analyze and Create ... to launch the SCI Module.

When starting the SCI module, three widows pop up: (1) the inform-

ation window about the plugin. This window can be closed imme-

diately. (2) the window where you can monitor/display the coloured

inverted data and compare to the original data during the design of the

operator. This window can be closed and re-open later with the

icon. (3) the SCI main window : this is the window where we are work-

ing. In the SCI  main  window, the main icons for the operator design

are: (used in the order from left to right, the two last

ones being optional in the workflow)

- 213 -

Step 2 - Selecting Input Data

In order to design the operator with the SCI application, it is first necessary toselect and analyze the seismic and well data spectra. This is achieved by loadingsome seismic trace data and well log impedance data (in time).

1. Open the Select Input data window with the icon

2. Selecting Seismic dataa. Click on Input Seismic and select 1-Original Seismic.b. Click  Load Seismic to load the default 40 traces randomly selected in

the defined survey area.c. Limit the vertical selection using a range or a horizon. Select the hori-

zon Demo 0 --> FS4 and toggle the Range to Horizon. The Intervallength should now display as 'Relative' and be -500ms to +500 ms.

- 214 -

3. Selecting Well dataa. Click on Input Well Logs tab.b. Click on Load Wells.c. Select several wells with AI or P-Impedance. Right-click on the well to

generate an Acoustic Impedance log if it is not loaded yet.d. To limit the vertical selection, the previously selected horizon should

be selected, toggle the Range to Horizon (Interval length shouldmatch that of the seismic, i.e 'Relative', - 500ms to +500 ms long).

- 215 -

4. Close the "Select Input Data" window.

Step 1 has filled the first, second and third rows of graphs. The other

graphs are affected by the rest of the workflow.

Step 3 - Design the Operator

Various parameters exist which allow you to perturb how the operator is gen-erated. These changes occur in real time so you will be able to see immediatelythe effect of the change you have made.

Optionally you can re-open the Seismic window with the      icon.

1. Open the Design Controls Dialog by clicking the       icon.

- 216 -

2. Smooth the amplitude-frequency plot of seismic data (seismic mean) bychanging the smoothing operator length.

Row 1

3. Smooth the amplitude-frequency plot of well data by changing the low/highcut (unselect Full Range).

- 217 -

Row 3

To change the scale used for one of the axis, right click on the graph

and select X axis or Y axis and select Logarithmic/Linear scale.

4. In the Design Operator section, tweak the parameters (low cut, high cut)(unselect Auto Calc.). The residual operator (blue curve) must stay  0 in thefrequency domain, with a quick drop on both sides. The effect of the para-meter tweaking is immediately visible on the seismic display that is updatedautomatically. Notice the seismic ringing that is introduced when the resid-ual operator is not flat in the low frequencies (Low cut parameter in the 0-8Hz range).

Row 4, Column 1

Change the high/low cut and the number of zero-crossings. Look at

- 218 -

the operator in the time domain and change the number of zero- cross-ings so the operator is cut where the amplitude starts to be negligible.

Row 3, Column 2

Observe the -90˚ phase of the operator.

If the  QC operator does not have the desired  shape, parameter set-tings in the seismic coloured inversion workflow must be changed,until the correct QC operator shape is established.

5. Save the operator by giving it a name. You can optionally save your

session as well.

Step 4 - Apply the Operator

1. Close the SCI main window.2. Back in the main Attribute Set window, select your operator as your input

wavelet (with the extension _ci).

The SCI operators and SSB operators are stored as wavelets.

- 219 -

3. Name the attribute and Add as new.4. Display the design operator on the inline 4255. Compare the result with the seismic data used as input and, if satisfied, cre-

ate a volume output.

Original Seismic

Coloured inversion

2.5.2   Absolute Impedance Inversion (DI & MPSI)

What you should know about DI and MPSI:

The Deterministic Inversion (DI) plug-in inverts the seismic data using an a prioriimpedance model. The output is an estimate of the mean impedance at eachsample location. The prior model is created first using stochastic parameters

- 220 -

(variograms) extracted from the data. Then a 2D error grid volume is constructedto get spatially variable constraints. Finally the model, error grid, seismic volumeand wavelet are used to create the mean impedance volume.

The MPSI (Multi-Point Stochastic Inversion) module starts after the deterministicinversion. Many realizations of the impedance inversion are computed startingfrom the mean impedance volume (from DI) using the stochastic model para-meters input in the a priori model building step, and a user-defined NScore trans-form. Several utilities can then be used to convert the  realizations into geobodies,or probability volumes.

Variogram Analysis

What you should know about variograms:

Variogram modeling is free (open source) in OpendTect. It is included

in the commercial part of this training manual because variogram para-

meters are necessary inputs for deterministic and stochastic inversion

described hereafter.

A variogram describes the spatial continuity of a variable. The inversion model inthe upcoming exercises will be constructed in three zones or layers bounded bytwo horizons. These horizons are represented in the wells by the FS8 and FS4markers.

Both horizontal and vertical variograms will be computed for the packages aboveFS8, between FS8 and FS4, and below FS4.

Horizontal semi-variograms

Horizontal variograms are computed from grids (attributes) stored on horizons.The attribute used for this analysis is the inversion target, impedance maps.Nevertheless one should not forget that stationarity is a basic assumption in vari-ogram analysis. Stationarity implies that the variograms analysis should be per-formed on trendless data. An average impedance map extracted from abroadband mean impedance volume is very unlikely to show no trends, thus it rep-resents an improper intput. The closest maps that can be produced, and that doesnot contain trend(s) are attribute maps extracted from relative impedancevolumes.

- 221 -

EXERCISE 2.5.2a

Objective:Extract horizontal variogram parameters from color inverted grids.

Workflow:

Data preparation

1. Run Seismic Coloured Inversion on the input seismic volume (see exercise2.5.1).

2. Map at least the top and/or the base of several layers by tracking the zero-crossing in the SCI volume. In this case use the horizon Demo 0 --> FS4.

l Extract an attribute map using:o Volume Statistics attribute: stepout set to 0, time gate, minimumor maximum),

o Event attribute: using the SCI volume as input, selectmultipleevents, use the event type maximum and output the amplitude.

o Stratal amplitude: Processing > Create Horizon Output >StratalAmplitude.

o Here use the Stratal amplitude and select the horizon Demo 0 --> FS4 and a time gate [-20,+20]ms and select maximum amp-litude.

3. Load the attribute at the horizon location by right-clicking > select HorizonData and select the one you just created.

4. Right-click on the loaded horizon data: follow Tools > Variogram... Vari-ogram analysis

This is an analysis tool that allows you to know about the lateral vari-

anility of your data. This information is in particular used when building

the model for the acoustic inversion.

5. You can change the maximum range (maximum distance allowed betweenthe pairs for the analysis). Each lag will be analyzed using a random subse-

- 222 -

lection of pairs. Keep the default values.

6. In the variograms window (see below) you can set the synthetic variogram(green) by setting its model, range and sill that bests fit your data (bluecurve). Mind the impact of the number of pairs per lag on the smoothness ofthe data extracted curve.

Examples of horizontal variograms from the Stratton field. The foldincreases from left to right, with respectively 1000, 10000 and 100000points per lag distance.

Vertical semi-variograms

Vertical variograms need to be extracted similarly. Although volume attributescould be used, well log measurements represent a more reliable input. The ver-tical variogram tool extracts its input from well logs using the Well vs. Attributecross-plot tool. The log data is resampled at the variogram processing step andde-trended prior to the variogram computation itself.The variogram analysis is performed in the time domain since the inversion is per-formed in this domain. As a result the wells used to extract the log data must beproperly tied before performing the variogram analysis.

- 223 -

EXERCISE 2.5.2b

Objective:Extract vertical variogram parameters from impedance well logs.

Workflow:

Data preparation

1. Extract P-Impedance logs using theWell vs. Attribute cross-plot module inthe Analysis > Cross-plot menu.

The data is extracted at the survey sampling rate. It is recommended to

lower the survey sampling rate to 1ms during the variogram analysis.

l Select the wells to be used for the data extraction and the P-associated impedance log.

l Set the “Radius around wells” to 0.l Choose “Nearest sample” as “Log resampling method”.

2. The extracted data will be shown in the cross-plot table window.

3. Select your P-Impedance column and press the variograms icon in thetoolbar.

Variogram analysis

The input parameters are very much comparable to the horizontal vari-

ograms analysis. The main difference is the number of available data

points. Variogram analysis requires a minimum number of pairs per lag

distance and lots of data must be input in order to obtain a rep-

resentative variogram.

The analysis can be performed well by well to get an idea of the vari-

ability, but it is advised to estimate the final variogram range from all

- 224 -

wells. If not enough data was collected you can either lower the min-

imum number of pairs or increase the size of the interval used for the

log extraction.

Examples of vertical variograms from the Stratton field. From left toright: above, in and below the target interval [mfrio C38]. The data wasextracted using an average filter (top) or the nearest sample (bottom).Note the impact of the filter on the variograms shape for the very firstlag distances.

Deterministic Inversion

The exercise regarding Deterministic Inversion is presented in the ded-

icated MPSI manual from Earthworks & ARK CLS (a separate doc-

ument). This manual can be found on opendtect.org website, in

Support > Documentation (link: http://www.opendtect.or-

g/pdfs/EWManual.pdf)

A few comments only will be given here.

- 225 -

Wells preparationIt is of uppermost importance to have  a fully prepared log database. There shouldbe no spikes in the logs, and it is recommended to have extended the logsupwards and downwards such that they cover the Z range where the a priorimodel will be used. Also since the model is created in the time domain the wellsmust be tied to the seismic before the inversion. Finally the logs must be in thesame units and it is preferable that each log is called with the same name in allwells.

OpendTect comes with an extensive Rock Physics library to create logs fromexisting logs. The rocks physics module is called from the Manage Wells utility by

pressing the Create button, followed by the Rock Physics icon . This icon isnow also located in the OpendTect toolbar (top of the screen).The utility supports creating new logs and filling in holes in existing logs. Anexample of the latter is given below where Gardner’s equation is used to replaceundefined values in the density log.

Some useful equations in the context of MPSI inversion are:l Gardner’s equation for computing density from a sonic log.l Castagna’s equation for computing Shear sonic from a sonic log.l Krief’s equation for computing Shear sonic from a sonic log.l Poisson’s Ratio.l Acoustic impedance from density and sonic.

- 226 -

l Elastic impedance from density and shear sonic.l Extended elastic impedance for the angle χ (in radians).

Wavelet polarityThe wavelet polarity is always subject to many mistakes and many confusions.The polarity of the wavelet in the MPSI deterministic inversion attribute should beset to Normal, and set the wavelet polarity using the wavelet manager. In the fig-ure below the wavelet on the left has a positive amplitude at t=0ms. This is a true(almost) zero phase wavelet. The wavelet on the right is the opposite, it has atrough at t=0ms, so it is a reversed zero phase wavelet.

Left: Zero phase wavelet. Convolving this wavelet with reflectivity series from thewells will associate peaks (+ve, positive amplitudes) in your seismic survey with alocal impedance increase.

Right: Reversed zero phase wavelet. The real phase of this wavelet is +/- 180degrees. Convolving this wavelet with reflectivity series from the wells will asso-ciate troughs (-ve, negative amplitudes) in your seismic survey with a local imped-ance increase.

The ideal workflow for setting the wavelet polarity should be the following:

1. Extract the statistical wavelet from the seismic survey. If you know the polar-ity of your dataset, set directly the phase to either 0 or 180. 0 will provide awavelet like the left example, 180 will create a wavelet like the rightexample.

- 227 -

2. Tie the wells to the seismic data. If the polarity is correctly set and the well istied with a good match, then the cross-correlation graph will show a largepositive value at zero lag. If the polarity is wrongly set you should see astrong negative amplitude at zero lag, or offset from the zero lag if the pre-vious interpreter tied the well assuming the wrong polarity.

3. Correct the polarity if needed in the wavelet manager using either theReverse polarity button or the Rotate phase button.

4. Apply the MPSI attribute with the polarity flag set to Normal. The option“Reverse” reverses the wavelet polarity on the fly without changing thestored wavelet on disk. This is not recommended.

Scaler extractionThe computation of the  scaler is the  most difficult part of the  impedance inver-sion, after the generation of the wavelet.

The following guidelines should always be honored:

l The scaler varies with the frequency content of the a priori model. Ideally thescaler should be computed on unfiltered a priori models. Thus all smoothingparameters from the 3Dmodel attribute should be toggled off during thescaler computation. Smoothing can be turned on again for running the inver-sion by setting the scaler to the computed value.

l The scaler should always be set with ‘relaxed’ constraints, set to 0.1, 0.1.l The scaler is computed over the Z range of the working area. The survey Zrange is thus far too large, and you must lower the Z range of the workingarea to your target area for the computation of the scaler. This option is avail-able in the View menu.

l The scaler is by default extracted along the wells. Sometimes this data isnot suitable for the scaler extraction, and one needs to compute the scalerfrom a subselection of the points. Note that both use the a priori model asinput for computing the synthetic seismic, and not the impedance logs fromthe wells.

If the scaler is too low: The inverted impedance will have very strong vertical vari-ations, and the corresponding synthetic seismic error will be very similar to theinput seismic volume, both in the time domain and in the frequency domain.If the scaler too high: The inverted impedance will be very similar to the a priorimodel, and the corresponding synthetic seismic error will show low overall amp-litudes.

LN Error correction

- 228 -

This option applies a lognormal correction to the AI log when converting the syn-thetic seismic error to impedance errors. The automatic computation trans-formation goes sometimes wrong and one must then toggle off this option ormanually set to value. A symptom to be kept in mind is that a wrongly set LN errorcorrection can shift the mean impedance value over the target.

Block sizeThe block size determines the blockiness of the output impedance. A large blocksize will have poor resolution but a better estimate of the mean impedance. Theblock size should ideally be set to 1. Using a larger value may increase the reli-ability of the mean impedance estimate, but will return a rather blocky output thatcan be more difficult to interpret.

Pre-processingThis function is used to reduce the runtime of the attribute. However the pre- pro-cessing must be re-done everytime the scaler, well constraints, block size or inver-sion range (Z range of working area) is changed. As a rule it should not be usedduring the testing phase, but after the parameters have been finalized and beforethe batch processing.Even so, for small targets it can be more efficient to compute the matrices on thefly than to read them from disk. An approximation is to use the pre-processingwhen the inversion window is larger than 500 samples.

Quality control of the inversion:

l QC that the synthetic seismic error has an RMS amplitude lower than 10%of the RMS of the seismic.

l When inverting with relaxed constraints, QC that the inverted impedance cor-relates with the impedance logs using the cross-plot tool. Optimize thecross-correlation as a function of the scaler and LN Error correction.

l Extract  the  histogram  from  the  inverted  impedance  volume  and meas-urement impedance log using the well-attrib cross-plot tool, around/in theinversion target. QC that the mean and standard deviation are similar. Ashift in the mean indciates an LN Error problem while a change in the stand-ard deviation is indicative of a problem with the scaler.

Stochastic Inversion

The exercise regarding Stochastic Inversion is presented in the ded-

icated MPSI manual from Earthworks & ARK CLS (a separate

- 229 -

document). This manual can be found on opendtect.org website, in

Support > Documentation (link: http://www.opendtect.or-

g/pdfs/EWManual.pdf)

The parameters to set for the stochastic inversion are rather limited. One mustenter low and high bounds for the stochastic impedance generation. However thestatistics shown in the log window are computed from the entire Z range, and arecertainly not fit for the purpose of the inversion.

Better minimum and maximum values should be extracted using the well-attribcross-plot module in the range of the impedance inversion. The histogram willreturn minimum, maximum, average and standard deviation values for the level ofinterest.

The figure above shows an example of a P-Impedance histogram extracted at theinversion level. The distribution is well behaved except for the high end of the his-togram, very much stretched towards large impedances. This could be the resultof the presence of many high impedance spikes. If not then it is still not recom-mended to use the maximum value as parameter for the stochastic inversion.

The lower and high bounds should be within +/-2 or maximum 3 standard devi-ations from the mean impedance value.

2.5.3   Porosity Prediction using Neural NetworksWhat you should know about Neural Networks for Rock Property Prediction

- 230 -

l This is a supervised approach using a Multi-Layer-Perceptron neural net-work.

l The network will find the optimal (non-linear) mapping between seismicattributes (usually impedance logs) and target well log attributes (porosity,gamma-ray, Vshale, Sw …).

l The network is trained on data points extracted along the well tracks.l Part of the extracted points are used as test set to determine the optimalpoint to stop training and avoid over fitting.

l The trained network is applied to (inverted) seismic data.l The input data (inverted seismic) needs to be scaled to match the scaling ofthe input data set that was used in training (logs).

Details:In the exercise that follows, we will convert seismic information to porosity using aneural network inversion workflow.

As in the chimney cube exercise (see Chapter Seismic Object Detection usingNeural Networks), we will use a supervised neural network to establish the (pos-sibly non-linear) relationship between seismic response and porosity. The main difference from the previous exercise  is that we will now use well information toconstruct the training (and test) sets.

The input consists of acoustic impedance values from the AI volume and the ref-erence time, i.e. the two-way time at the extraction point. The reference time isincluded to capture a possible porosity trend with depth (time).

Theoretically we only need the AI value at the evaluation point as input to theneural network but this assumes that the inversion process has completelyremoved the wavelet and that there is perfect alignment of AI and log responsesalong the entire well track. To compensate for potential inaccuracies we willextract more than just the AI value at the evaluation point. Instead we will extractAI in a 24ms time window that slides along the well tracks. The correspondingporosity values from the depth-to-time converted and resampled logs serve as tar-get values for the neural network.

Porosity prediction is a relatively easy process. The workflow is schematicallyshown below:

- 231 -

Log property prediction workflow

This workflow can be used to create log property cubes such as a Porosity Cubeand a Vshale Cube.

EXERCISE 2.5.3

Objective:Predict porosity from inverted acoustic impedance and well logs using a neuralnetwork.

Workflow:

Step 1 – Data preparation

Well data

1. Well(s) need to be available in the survey. If they are not there available:import wells (track, logs, markersand optionally time-depth curve or check-shot).

2. Tie well to the seismic (see exercise 1.6.1)

It is very important to have a good well tie for this worklow is performed

in the time domain.

- 232 -

Attributes

3. Open the Attribute Set window by clicking on the icon.

4. Open the saved attribute set called Inversion attributes from the icon.

Reference time is defined in the attribute set.

In theory, only the acoustic impedance value at the evaluation sample

is needed as input for the neural network analysis. However this

assumes that the inversion process completely removed the wavelet

and that the acoustic impedance volume and the log response are per-

fectly aligned. To compensate for possible inaccuracies, the amplitude

from the acoustic impedance cube is extracted in a time window

(defined regarding the evaluation sample) that slides along the well

tracks.

The amplitudes from the AI cube are extracted in a time gate of -12 to +12ms rel-ative to the evaluation point.

Click Close to dismiss the window.

Step 2 – Neural Network data selection

5. Open the neural network plug-in with the icon.6. Select Property prediction [Well Data]... in the Neural Network manager.

- 233 -

Neural Network manager

7. The data selection window pops up.

8. Select the Input Attributes

- 234 -

All the attributes from the active attribute set are listed here with the

stored attributes (name written between [ ]).

9. Select your Target log: Porosity. This is the property you want to predict.This log need to be defined in at least one well in your interval of interest.

10. Select all 4 available wells (F02-1, F03-2, F03-4 and F06-1).

- 235 -

11. Define the area and interval of data extraction:l Vertically: limit the extraction of data to your interval of interest usingthe well markers +/- some additional distance. Extract between topFS8 and FS4 (distance above and below are both 0).

l Laterally: the data can be extracted along traces within a given radiusaround the well. To extract only at the well location, set the distance to0.

12. Select the Vertical sampling method: Average. This means that the targetporosity is calculated over all  well log values within a window of + and -half a sample rate.

13. Additional information may need be provided:l Choose the type of property to be predicted:Ordinary well log values.l The extracted data set is split randomly into a training and a testingdata set.

For a usual log with varying values, select the “Ordinary well log val-

ues”. In case of a log containing only discrete values such as a litho-

log, select “Lithology codes”.

You want to have enough data in the train dataset and in the test data-

set.

14. Click OK.

Step 3 – Neural Network processing

15. The data is extracted and the two dataset (training and testing) are created.The data are presented in a table (similar to a crossplot table).

- 236 -

Optional  QC Step: To  cross-plot  Acoustic  Impedance  versusPorosity (figure below)

a. click on the Header of the Porosity column and press the X icon.b. click on the AI 0 column, and press Y+ followed by the Cross-

plot icon. The cross-plot window will popup.c. Use the green arrows (in the previous window) to move the Y

column  to  see how the other input attributes plot against poros-ity.

d. Press the Distribution icon to see how the data are distributed.The cross-plotting utility also allows us to edit the data. Forexample,  we  can  remove  outliers,  which  will  improve sub-sequent neural network training.

e. In this case, no editing is needed, so dismiss the cross-plot win-dow and press Run to continue.

Porosity versus AI cross-plot

16. Press Run.

You have now the option to balance your data. Balancing is a recom-mended pre-processing step  (see figure below). It  helps the networkto find  the relationship we  seek instead  of finding  a relationship thatis driven by the sample distribution.

- 237 -

Data preparation of the training dataset.

The aim of balancing is to give the network equal amounts of training

samples over the entire range.

Imagine that the range of porosities varies between 2% and 30% butthat 90% of the data has porosities between for example 18% and20%. The trained network will then probably also predict only valuesbetween 18% and 20% as such a network will have the smallest pre-diction error over all the samples in the training set.

Balancing is done by dividing the distribution chart into a user-definednumber of classes. The desired number of classes is set. Classes thatare over-sampled are re-sampled by randomly removing samples. Thenumber of samples in  under-sampled classes is increased by ran-domly selecting a sample, adding 1% of white noise and adding thenoisy sample to the training set. This process is repeated until allclasses have the same number of samples.

In this case we are lucky that our dataset is nicely sampled as wehave learned in the previous cross-plotting phase.

- 238 -

To continue, press Ok.

17. The network training now starts automatically.

Neural Network training performance window

The cross-plot in lower-left corner plots predicted target values againstactual values for train and test sets. For an explanation of the errorcurves and the color scheme of the input nodes see the previous exer-cises.

Stop training when the error on the test set is minimal (see below).

If you are not sure if you are at the minimum or you over-

trained your neural network, you can “clear” the neural net-

work and then the neural network is re-trained.

- 239 -

Press Ok and Save as with an appropriate name.

Step 4 – Neural Network results: display and QC

18. After stopping the neural network process, a report is created summarizingthe neural network characteristics.

Neural Network Report

Results QC

The Neural Network results need to be QCed at well location.

19. Add an inline in the scene at well location: select the well F03-4.20. Add an attribute: select Porosity in the Neural Network category.21. Create a low-pass filtered version of the Porosity log for the well F03-4.

When comparing seismic and well data, you should take into account

the difference in frequency content and thus you should apply a low-

- 240 -

pass filter (depending on the spectrum of your seismic) on your log

before using it.

l Open the well manager .

l Click on below the well list.l Select the well F03-4 and the log Porosity. Use thedefault values for the other parameters.

l Click on Continue.l In the pop-up window, the original log is displayed inblue.

l In the actions listed below, select: FFT filter.l Choose the filter type: LowPass.l Set the maximum frequency to 70Hzl Click Apply: the filtered log appears in red.l Optionally change the extension to add to the logname to _LP70Hz.

l Click OKl The log has been created: close the well manager.

22. Load the well F03-4 and display the log Porosity_LP70Hz.23. Adjust the color bar to have the same for the log and for the attribute.

QC at the well location

- 241 -

Porosity prediction result on random line through 4 wells. The dis-played log is the actual porosity trace.

24. If you are satisfied with the results you can then proceed with creating theporosity volume: Processing > Create Seismic output > Single Attribute >3D... and in the Neural Network category, select the property: Porosity.

References

De Groot, P., 1999 Seismic Reservoir Characterisation Using Artificial NeuralNetworks. 19th Mintrop seminar, Muenster, Germany

Berge, T.B., Aminzadeh, F., de Groot, P. and Oldenziel, T., 2002 Seismic inver-sion successfully predicts reservoir, porosity, and gas content in Ibhubesi Field,Orange Basin, South Africa. The Leading Edge, April

- 242 -

Appendix - GMT SoftwareGeneric Mapping Tools (GMT)

GMT is an open source collection of tools for manipulatinggeographic and Cartesian data sets and producing Encapsulated Postscript (eps)file illustrations ranging from simple x-y plots via contour maps to artificially illu-minated surfaces and 3-D perspectives views.

In this appendix, we will shortly explain the GMT plug-in and we willcreate different maps in OpendTect:

To launch GMT tools, click on the icon in OpendTect main toolbar. The first timeyou launch the GMT mapping tools, a warning message will pop-up: a mappingtool package needs to be installed in order to run it. This can be downloaded fromthe GMT web site: http://gmt.soest.hawaii.edu/projects/gmt/wiki

If OpendTect fails to create a map with GMT, check whether the envir-

onment variable GMTROOT is set to the directory in which GMT was

installed and whether the PATH variable includes the GMT bin dir-

ectory. (Per default: GMTROOT c:\programs\GMT4 and PATH …c:\-

programs\GMT4\bin…). Environment variables in Windows 7 can be

set from Computer > System Properties > Advanced System Settings.

After successful installation of GMT package, the GMT user interface will be star-ted:

- 243 -

When creating postscript maps, the several tabs allow to specify the respectivesettings:

l Basemap: used to set the scale of map and other map settings.You do not need to add it in the map overlays. This is the firstand mandatory step into the creation of maps

l Locations: used to post pickset data (e.g. proposed well loc-ations) in the map overlay

l Polyline: used to add polygons (e.g. lease boundaries) in themap overlay

l Contours: used to make a horizon contour mapl Coastline: used to draw coastal linesl Wells: used to post wells in the mapl 2D Lines: used to post the OpendTect 2D-Line(s) in the mapl Random Lines: used to post the Random Line(s) in the mapl Clipping: used to set up polygonal clip pathl Advanced: used to customize the GMT commands

- 244 -

A typical example of a time Contour Map with well locations

- 245 -

Personal Notes

- 246 -