Property Modeling

Course

Petrel 2010
About Petrel*
Development on Petrel seismic-to-simulation software began in 1996
in an attempt to combat the growing trend of increasingly specialized
geoscientists working in increasing isolation. The result was an
integrated workflow tool that allows E&P companies to think critically
and creatively about their reservoir modeling procedures and enables
specialized geoscientists to work together seamlessly. With the
enhanced geophysical tools and the integration of ECLIPSE* reservoir
simulation software and streamline simulation, Petrel is now a
complete seismic-to-simulation application for
• 3D visualization
• 3D mapping
• 3D and 2D seismic interpretation
• well correlation
• 3D grid design for geology and reservoir simulation
• depth conversion
• 3D reservoir modeling
• 3D well design
• upscaling
• volume calculation
• plotting
• post processing
• streamline simulation
• ECLIPSE
Copyright Notice
© 1998-2010 Schlumberger. All rights reserved.
No part of this manual may be reproduced, stored in a retrieval system,
or translated in any form or by any means, electronic or mechanical,
including photocopying and recording, without the prior written
permission of Schlumberger Information Solutions, 5599 San Felipe,
Suite 1700, Houston, TX 77056-2722.

Disclaimer
Use of this product is governed by the License Agreement.
Schlumberger makes no warranties, express, implied, or statutory, with
respect to the product described herein and disclaims without limitation
any warranties of merchantability or fitness for a particular purpose.
Schlumberger reserves the right to revise the information in this manual
at any time without notice.

Trademark Information
*Mark of Schlumberger. Certain other products and product names are
trademarks or registered trademarks of their respective companies or
organizations.
Table of Contents
About Petrel* ...........................................................................2
Copyright Notice........................................................................3
Disclaimer..................................................................................3
Trademark Information..............................................................3

Module 1 - Introduction...........................................................13
Prerequisites............................................................................14
Learning Objectives.................................................................14
What to Expect........................................................................15
Icons........................................................................................16
Lesson......................................................................................17
Course outline.........................................................................18
Property modeling workflow...................................................23
Terminology.............................................................................24
Acronyms used in this manual:...............................................27

Module 2 - Basic Statistics....................................................29

Lesson......................................................................................30
Geostatistics............................................................................30
Histogram................................................................................34
Theoretical distributions.........................................................37
Normal Score Transformation ................................................42
Crossplot and Correlation coefficient.....................................43
Generic Filters.........................................................................45
Basic Statistics Part 1 – Exercises..........................................46
Display and analyze a Histogram and a CDF curve.................46
Create a Normal Distribution from a Histogram.....................48
Display a Crossplot and calculate the Correlation
Coefficient...............................................................................49
Lesson......................................................................................52
Variograms...............................................................................52
How is a Variogram computed?..............................................54
Variogram Model Types...........................................................57
Why do we need Variograms?................................................61
The Variogram ‘search cone’...................................................64
Sample variogram theory........................................................69
Basic Statistics Part 2 – Exercises..........................................73
Variogram map calculation of a point data set . ....................73
Sample variogram calculation of point data set.....................75
Property Modeling Table of Contents • 5
 Define a variogram model.......................................................76
Summary..................................................................................78

Module 3 - Kriging....................................................................79
Lesson......................................................................................80
Kriging.....................................................................................80
Kriging types............................................................................84
Variogram model type.............................................................89
Variogram range......................................................................90
Variogram azimuth...................................................................91
Variogram nugget....................................................................92
Kriging – Exercises..................................................................94
Influence of the variogram model parameters on Kriging
results......................................................................................94
Summary..................................................................................97

Module 4 - Gaussian Simulation...........................................99

Lesson....................................................................................100
Gaussian simulation vs. Kriging............................................100
Why Gaussian simulation?....................................................102
Change of Support Effect .....................................................109
Unconditional simulation .....................................................113
Gaussian Simulation – Exercises..........................................116
Influence of variogram model parameters on Gaussian
Simulation.............................................................................116
Optional Exercise: User defined Normal distribution for
simulation..............................................................................118
Summary................................................................................120

Module 5 - 3D Property Modeling Overview.....................121

Lesson....................................................................................122
Petrel workflow tools............................................................122
Facies modeling.....................................................................124
Deterministic algorithms . ....................................................124
Stochastic algorithms ..........................................................126
Petrophysical modeling.........................................................128
Deterministic techniques......................................................128
Stochastic techniques...........................................................130
Summary................................................................................132

6 • Table of Contents Property Modeling

 Module 6 - Property Modeling Data Preparation.............133
Lesson....................................................................................134
Reservoir modeling ..............................................................134
Facies / Lithology Interpretation...........................................136
Artificial Neural Networks....................................................139
Petrophysical log interpretation............................................144
Property Modeling Data Preparation – Exercises.................151
Facies calculation..................................................................151
Facies interactive interpretation...........................................152
Neural Networks Classification............................................153
Optional: Neural Networks to Make Well Logs....................158
Create a Water Saturation (SW) Log and Property...............159
Summary................................................................................165

Module 7 - Scale Up Well Logs...........................................167

Lesson....................................................................................168
Scale up well logs - Principle................................................168
Scale up Discrete well logs .................................................170
Scale up Continuous well logs..............................................172
Averaging methods...............................................................172
Bias to a discrete log.............................................................173
Cells size in I,J,K direction....................................................178
Averaging logs from multiple wells......................................179
Saved searches.....................................................................181
Scale up well logs – Quality Control.....................................182
Filters.....................................................................................184
Scale up well logs – Exercises..............................................184
Scale up discrete well logs...................................................184
Scale up continuous well logs..............................................186
Optional: Quality Control of upscaled logs...........................189
Summary................................................................................191

Module 8 - 3D Grid Quality Control Tools...........................193

Lesson....................................................................................194
Univariate Statistics (Discrete properties) . .........................195
Univariate Statistics (Continuous properties) ......................196
Property Filters......................................................................197
Why is Correlation important?..............................................199
Crossplot................................................................................200
Variogram analysis ...............................................................201
Property Modeling Table of Contents • 7
 Simbox mode.........................................................................202
3D Grid Right sizing...............................................................203
3D Grid Quality Control Tools – Exercises............................206
QC of 3D Grid layer resolution .............................................206
Summary................................................................................214

Module 9 - Facies Modeling................................................215

Lesson....................................................................................216
Facies Modeling....................................................................216
Carbonate depositional environments..................................218
Clastic depositional environments........................................219
How to build a facies model?................................................221
Facies modeling methods in Petrel.......................................223
Data Analysis - Facies modeling...........................................227
Facies Data Analysis – Exercises..........................................235
Vertical facies proportion analysis........................................235
Facies thickness analysis......................................................236
Facies probability analysis....................................................238
Variogram analysis................................................................239
Summary................................................................................242

Module 10 - Sequential Indicator Simulation...................243

Introduction...........................................................................243
Facies modeling methods in Petrel.......................................244
Sequential indicator simulation (SIS) ..................................245
Sequential Indicator Simulation - Theory.............................247
Sequential Indicator Simulation - Setup in Petrel................248
Global fraction.......................................................................250
How to generate Trend input in Petrel..................................251
Sequential Indicator Simulation – Exercises........................253
Influence of Variogram parameters on a Facies model.........253
Influence of ‘Global facies proportion’ on Facies
modeling................................................................................254
Using Variograms and Trends for Facies model
generation.............................................................................258
Sequential Indicator Simulation with seismic attribute.......261
Summary................................................................................263

Module 11 - Object Facies Modeling.................................265

Lesson....................................................................................266
8 • Table of Contents Property Modeling
 Facies modeling methods available in Petrel.......................266
Object modeling overview.....................................................267
Adaptive channel modeling...................................................268
Object modeling – process....................................................278
Hierarchical modeling...........................................................281
Object Facies Modeling – Exercises.....................................282
Adaptive channel modeling - Zone A....................................282
Optional Exercise: Adaptive channel modeling with
trends.....................................................................................288
Adaptive channel modeling using body index property........290
Optional Exercise: Hierarchical facies modeling..................295
Summary................................................................................298

Module 12 - Truncated Gaussian Simulation....................299

Lesson....................................................................................300
Truncated Gaussian simulation (TGS)...................................301
Theory - How does TGS work?..............................................302
Facies associations...............................................................304
Variograms.............................................................................305
Truncated Gaussian with trends...........................................306
Geometry tab (Trend model)..................................................311
Aggradation angle.................................................................312
Edit transition lines...............................................................313
Truncated Gaussian Simulation - Exercises..........................314
Modeling a transitional depositional environment
using the “Truncated Gaussian simulation” method............314
Modeling a transitional depositional environment
using the ‘Truncated Gaussian with trends’ method............316
Summary................................................................................320

Module 13 - Petrophysical Modeling.................................321

Lesson . .................................................................................322
Petrophysical modeling - overview.......................................322
Modeling input......................................................................324
Data analysis - Petrophysical modeling................................325
Data analysis - Input Distribution.........................................326
Transformations.....................................................................327
Spatial trends........................................................................332
Example of a transformations sequence...............................334
Variograms.............................................................................335
Property Modeling Table of Contents • 9
 Variogram search cone..........................................................337
Variogram modeling process.................................................338
Data Analysis – Exercises.....................................................340
Data Transformation..............................................................340
Applying Normal Score Transformation................................341
Applying Trend Transformation.............................................343
Variograms.............................................................................347
Vertical and directional variogram analysis for porosity......347
Summary................................................................................352

Module 14 - Gaussian Simulation in Petrophysical

modeling...................................................................................353
Lesson....................................................................................354
Normal score transformation................................................356
Petrophysical modeling – Zone settings tab.........................356
Petrophysical modeling – Common settings tab..................358
Common settings tab - Local model update.........................359
Petrophysical modeling using secondary data......................360
Correlation.............................................................................361
Variogram analysis................................................................362
Horizontal variogram map.....................................................364
Sample Variograms...............................................................364
Gaussian Simulation in Petrophysical modeling–
Exercises................................................................................365
Porosity modeling based on a Facies model using
Sequential Gaussian simulation method..............................366
Optional Exercise: Net-to-Gross modeling using Gaussian
random function simulation method.....................................370
Optional Exercise: Correlation with secondary data.............371
Variogram analysis using correlated attributes....................373
Creating a variogram map from an Acoustic Impedance
property.................................................................................373
Creating sample variogram from an Acoustic Impedance
property.................................................................................376
Creating a variogram model based on the sample
variogram...............................................................................378
Local model update of a Porosity model...............................381
Summary................................................................................384

10 • Table of Contents Property Modeling

 Module 15 - Kriging in Petrophysical Modeling..............385
Lesson....................................................................................386
Co-kriging..............................................................................388
Kriging Expert settings..........................................................390
Kriging In Petrophysical Modeling - Exercises......................390
Modeling a SW property using Kriging method and
secondary information .........................................................390
Summary................................................................................397

Module 16 - Petrophysical Modeling using

Secondary Data .....................................................................399
Lesson....................................................................................400
Local Varying Mean (LVM).....................................................401
What is Co-kriging and why do we use it?...........................403
Collocated co-kriging.............................................................404
Collocated co-kriging characteristics....................................404
3D trend.................................................................................407
Bivariate distributions...........................................................408
Petrophysical Modeling Using Secondary Data –
Exercises................................................................................411
Porosity modeling using a Seismic Attribute as
secondary variable................................................................411
Permeability modeling...........................................................413
Permeability modeling conditioned to a Facies model
and using Collocated co-kriging with Porosity......................414
Permeability modeling using Collocated co-kriging
with Porosity conditioned to a Seismic Attribute.................417
Permeability modeling using the Porosity trend...................419
Bivariate distribution - Crossplot bins...................................421
Permeability modeling using bivariate distribution..............423
Summary................................................................................427

Module 17 - Volume Calculation and Uncertainty

Analysis....................................................................................429
Lesson....................................................................................430
Volume Calculation................................................................431
Why Uncertainty?..................................................................433
Uncertainty principle in Petrel...............................................434
Uncertainty results and display.............................................437
Volume Calculation & Uncertainty Analysis - Exercises.......439
Property Modeling Table of Contents • 11
 Volume calculation using the property model.......................439
Uncertainty analysis using the volume calculation case......441
Use of realizations ...............................................................445
Summary................................................................................446

Module 18 - Case Study . ....................................................447

Porosity Modeling using an Acoustic Impedance cube as
secondary information...........................................................447
Lesson....................................................................................448
Porosity Modeling using an Acoustic Impedance Cube
as Secondary Information - Exercises...................................454
Calculate an acoustic impedance log for each well.............455
Cross plot impedance log against porosity log.....................456
Check the model resolution...................................................457
Create layers with thickness derived from data analysis.....460
Resample the property logs and acoustic impedance
cube into the model...............................................................461
Cross plot the resampled seismic cube with the
impedance upscaled log........................................................462
Derive the variogram model parameters from the
acoustic impedance property................................................464
Porosity modeling based on seismic acoustic impedance....470
Summary................................................................................472

References . ............................................................................473

Index.........................................................................................475

12 • Table of Contents Property Modeling

Module 1 - Introduction

The purpose of this course is to provide an understanding of how to

perform Facies and Petrophysical modeling in Petrel. The target is to
use all geological information available to build a realistic property
model. Property modeling is the process of filling the grid cells with
discrete or continuous properties.
In order to construct a proper model based on a 3D grid, it is important
to understand the basic concepts, algorithms and methods leading to
these results. Therefore, the first part of the course will focus on basic
geostatistical concepts like variograms, kriging and simulation, and will
also test out different methods to see both the benefits and the
limitations.
Property modeling in Petrel is split into three separate processes:
Geometrical, Facies and Petrophysical modeling. In addition, there are
other process steps which can be used when modeling properties:
Scale up well logs, Data analysis, Train Estimation Model (Neural Net),
Trend modeling, User-defined object creation, Training image and patern
creation and Fault Analysis (the last four processes will not be covered
in this course).
By attending this course, you will obtain an introduction to the
geostatistical and Property modeling functionalities in Petrel. Upon
completion of the course, you will have been introduced to enough
options in Petrel to be able to build a property model on your own. Due
to the scope of the software, there is a wide range of functionalities in
Petrel that will not be introduced in this course. Refer to our other
courses for more details on separate modules.

Property Modeling Introduction • 13

 Prerequisites
To successfully complete this course, the user must have knowledge of
the following:
• English Proficiency
• Basic Windows and practical computing skills
• Familiarity with Geology Fundamentals
• Familiarity with Geostatistical Fundamentals
• Familiarity with Reservoir Modeling
• Familiarity with Basic Property Modeling
• Petrel Introduction

Learning Objectives
The purpose of this course is to give the participant an understanding of
how to perform Property modeling in Petrel. The basic concepts and
algorithms of Property modeling will be explained together with the
software functions. The course focuses on the practical use of Petrel
functionality to perform property analysis and population. At the
completion of this course, you will be able to:
• Understand Basic Geostatistics Fundamentals
• Prepare data using Well correlation and Calculator tools
• Scale up well logs and quality check the resulting properties
• Perform both statistical discrete and continuous data analysis,
including how to transform data and how to create variograms
from data
• Run different Geostatistical Methods (algorithms) to populate
the model with data including geometrical properties, facies
and petrophysical properties
• Use the Facies modeling process to create both pixel-based
and object-based 3D models
• Use the Petrophysical modeling process to create different
petrophysical models of porosity, water saturation, net to gross
and permeability
• Use different output properties to correlate for possible
Co-kriging/Simulation
• Use different methods of quality checking both the input data
and the resulting model
• Run basic Volumetrics and Uncertainty on the generated
property models.
14 • Introduction  Property Modeling
What You Will Need
You will need the following hardware and applications in order to
perform the workflow:
• A personal computer with a minimum of 2GB of RAM; however,
we recommend 16GB of RAM for optimal performance.
• For Petrel 32-bit: Microsoft XP 32. For Petrel 64-bit: Vista 64 and
XP 64.
• A graphic card compatible with Petrel
• A Petrel license and license key
• Petrel Seismic to Simulation Software with the latest updates
• Training datasets

What to Expect
In this training material, you will encounter the following:
• Overview of each module
• Prerequisites to the module (if necessary)
• Learning objectives
• A workflow component
• Lesson(s)
• Scenario-based exercises
• You will also encounter notes, tips and best practices

Property Modeling Introduction • 15

 Icons
Throughout this manual, you will find icons in the margin representing
various kinds of information. These icons serve as at-a-glance
reminders of their associated text. See below for descriptions of what
each icon means.

Tips Notes Best Practices

This icon points you to a tip This icon indicates that the This icon indicates the best
that will make your work following information is way to perform a given
easier. particularly important. task when different options
are available.

Lessons
Warnings Questions
This icon identifies a
This icon indicates when This icon identifies the lesson, which covers a
you need to proceed with questions at the end of particular topic.
extreme caution. each lesson.

Prerequisites

This icon identifies any

Procedures
prerequisites that are Exercise Review Questions
required for the course, or
This icon identifies the This icon indicates that it’s This icon identifies the
for individual modules.
steps required to perform a your turn to practice the review questions at the
given task. procedure. end of each module.
.
you need to proceed with
extreme caution.

Prerequisites Learning Objectives What you will need

This This
This icon identifiesany
icon identifies any This icon indicates
This icon identifiesany
any
This icon identifies
identifiesanyany
prerequisites thatare
are learning objective set out applications, hardware,
prerequisites that learning objectives set out applications, hardware,
required for the
thecourse,
course,oror for the course, or for the datasets, or other material
required for for the course, or for the datasets,for
or other material
for current module. required the course.
for individual modules.
individual modules. current module. required for the course.
.

16 • Introduction  Property Modeling

Lesson

Petrel 2010
Property Modeling – Intermediate Level Course
Intro Geostatistics Property modeling Scale up well logs Data quality control
data preparation
Basic Statistics Kriging
Petrel Property modeling

Vs.
Objective and workflow

Gaussian Simulation

Volume
Facies Modeling Petrophysical Modeling Calculation

Discrete Data Stochastic Facies Countinuous Stochastic & Deterministic Use of secondary
analysis modeling Data analysis Petrophysical modeling information for
property modeling

Property Modeling
Course Outline

Day 1 Day 2 Day 3

Fundamentals of Geostatistics Facies Modeling Petrophysical Modeling
Introduction
Facies Modeling Introduction Petrophysical Modeling Introduction
Basic Statistics
Data Analysis (discrete properties) Data Analysis (continuous properties)
Kriging
Gaussian Simulation Sequential Indicator Simulation Gaussian Simulation in Petrophysical
Object Facies Modeling modeling
Input data preparation and QC Truncated Gaussian simulation Kriging in Petrophysical modeling
3D Property Modeling overview Petrophysical Modeling Using Secondary
Property Modeling Data Preparation Data
Quality Control Tools Optional modules:
Scale Up Well Logs -Volume Calculation / Uncertainty Analysis
-Case Study (Petrophysical Modeling Using
Seismic)

Property Modeling Introduction • 17

 Course Outline
This course will focus on:
Basic geostatistical concepts, such as variograms, Kriging and
Gaussian Simulation methods and will also demonstrate different
methods to illustrate both the benefits and the limitations of each
concept.
During the course you will learn how to prepare the well data using
different processes, such as Well correlation, the Wells calculator and
Neural Networks.
By using the Scale up well logs process, logs will be up-scaled into
the 3D grid cells that are penetrated by wells.
Different methods of quality checking both the input data and the
resulting model will be introduced during this course.
Before different population methods of discrete data (facies,
lithofacies, etc) are introduced, the statistical discrete data analysis will
be performed by using the Data analysis process.
Only the stochastic methods for Facies modeling will be covered (SIS,
TGS and Object modeling) in this course. In addition, the process of
Hierarchical Facies modeling will be shown.
The statistical continuous data analysis will be performed by using
the Data analysis process. This includes how to transform data and
how to create variograms from data.
In Petrophysical modeling, the stochastics (SGS and GRFS) and
estimation (Kriging) methods will be covered. The Petrophysical
modeling process will be focused on how to create different models of
porosity, permeability, water saturation and net to gross.
In addition, how to use secondary information, such as seismic data
or other properties for Petrophysical modeling by using the different
methods like Local varying mean, Co-kriging and Bivariate distribution,
will be covered.
At the end of the course, an optional topic for volume calculation is
presented to illustrate how to use the modeled properties to quantify
the hydrocarbons in the reservoir, as well as how to perform
uncertainty analysis using the volume calculation as a base case.

18 • Introduction  Property Modeling

 Property Modeling
Course objectives

Understand the 3D Property modeling workflow

Explain the geostatistical concepts for data analysis and modeling
techniques
Describe how to properly prepare reservoir data for Property
modeling (facies and petrophysics)
Capture, understand and use geological features to build a realistic
Facies Model
Populate Petrophysics conditioned to a high quality Facies model or
constrained to Seismic and/or Trend data
Demonstrate how to run a simple Volume calculation using various
property models and how to handle basic Uncertainty

The purpose of this course is to give the participant an understanding of

how to perform Property Modeling in Petrel. The basic concepts and
algorithms of the Property modeling process will be explained together This course is not
with the software functions. designed to teach theory on
geostatistics, but it contains
The course focuses on the practical use of Petrel functionality to some basic concepts to
facilitate the novice with
perform property analysis and population. some tools to start advanced
Property modeling in Petrel.

Property Modeling Introduction • 19

 Property Modeling
Introduction

Why Create a Reservoir Model?

Maximize the information usage to

optimize production
Reservoir properties are critical factors
affecting production
Calculate hydrocarbon volume in
place
Assist in the field development
Assess uncertainty

We are making big decisions based on limited data!

Reservoir models
When you build reservoir property models, the goal is to maximize the
value of data by incorporating all available information into a
quantitative digital representation.
A 3D geological model can:
• Handle large amounts of data
• Provide consistent analysis in 3D
• Give direct numerical input to flow simulation and pore volume
calculation
• Test and visualize multiple geological interpretations
• Assess uncertainty
The importance of modeling or distributing well log information in 3D
space should not be underestimated. 3D distribution of property data is
a powerful tool for understanding the spatial distribution based on your
hard well data and trends. 3D models lead to better assessments than
a set of 2D interpretations.

20 • Introduction  Property Modeling

 Property Modeling
Petrel Reservoir Modeling Workflow
Data preparation
and interpretation

Well Structural modeling

design

Upscaling & Facies

Post-process modeling

Petrophysical
Uncertainty modeling
analysis

Volume
Calculation

Petrel supports the entire Reservoir modeling workflow:

• Data import, preparation and interpretation through structural
modeling and property modeling (both facies and petrophysical
modeling).
• Model interrogation (e.g., volumetrics and uncertainty
analysis).
• Upscaling properties to Simulation Grids and fluid flow
simulation (with ECLIPSE and Streamline Simulation integrated
in Petrel).
An advantage of Petrel is that the software is fast and processing can
easily be repeated as new data is acquired, using the Workflow
editor.
Unlike applications in the past, Petrel offers an environment in which
the modeled results can evolve as new data and hypotheses become
available.
Property Modeling Introduction • 21
 Property Modeling
Data Integration
Input Data
Well data, seismic, production, previous geological studies
Reservoir Modeling >> Geostatistics

Geology Geophysics Petrophysics Engineering

Deterministic
Information
Integrated Study (framework)

Conceptual Statistical
Integrate maximum Information Information
amount of information (connectivity) (variation)

It is important to incorporate all available information.

To build realistic property models, you need input from different
disciplines, including Geophysics, Geology, Petrophysics and Reservoir
engineering. When building a model, it is crucial to integrate all
available data types.
In projects with sparse data:
1. The deterministic information gives the main framework
(structural-stratigraphical model)
2. The statistical information, such as histograms, variograms,
correlation and trends, gives the property variation in the
model.
3. The conceptual information is very important for the
following reasons:
• The use of analogs (outcrops, neighboring fields, etc.) may
serve as a source of statistical and geometrical
information.
22 • Introduction  Property Modeling
 • A combination of conceptual and statistical information
may allow the user to choose key parameter values and to
indicate upper and lower bounds for such key parameters
when little data exists in the project area (e.g. object
orientation, size, shape, correlation lengths and so on)

Property modeling workflow

Property modeling is the process of filling cells of the grid with
discrete or continuous properties. The target is to use all geological
information available to build a realistic property model.
Property modeling in Petrel is split into three separate processes:
- Geometrical modeling
- Facies modeling
- Petrophysical modeling
In addition, there are other process steps which can be used when
modeling properties:
Property Modeling Introduction • 23
 - Scale up well logs
- Data analysis
- Trend modeling (not covered in this course)
- User-defined object creation (not covered in this course)
- Training image and pattern creation (not covered in this course)
- Fault analysis (not covered in this course)
- Train estimation model (Artificial Neural Networks).

Terminology
Anisotropy – A way to measure whether variance within a collection
of data is determined by direction (measured in azimuth and percent
eccentricity).
Automatic Legend – A predefined template displaying the color table
legend of a displayed object.
Continuous– Digit number for a property with unlimited possible
options, e.g. porosity, permeability, etc.
Corner point grid – A flexible grid structure where the eight corners of
a cell (the nodes) can be moved to form irregular cell geometries.
Correlation – A way to measure whether two separate collections are
related (measured in percent).
Crossplot– Graphical representation of the values of two variables
measured at the same location and reveals the degree of correlation
between the two variables by the shape of the data cloud.
Data Analysis – The process of applying transformations on input data
(normally upscaled well logs) identifying trends and defining variograms
describing the data. This is then used in the facies and petrophysical
modeling to ensure that the same trends appear in the result.
Deterministic Algorithms – Based on a given input data set and the
final result will be calculated on a predefined formula. The output result
is unique and depends only on the data.
Discrete– Integer code for a property with a limited (countable)
number of possible options; e.g. facies, bodies or lithologies.

24 • Introduction  Property Modeling

Facies Modeling – Interpolation or simulation of discrete data, for
example facies.
Fault Analysis – The process where the user can generate fault
transmissibility multipliers, either directly or by modeling fault
properties, providing grid permeabilities and calculating the multiplier.
These are then used as input to the simulation or simply as a visual
assessment of the sealing potential of faults.
Geometrical Modeling – No interpolation of input data is required.
Properties are built based on the geometrical properties of the grid cells
themselves, distance to other objects etc., some algorithms also require
input data, but this data is simply sampled into the grid (e.g. seismic).
Histogram– Graphical representation of the frequency distribution of
the selected variable(s).
Horizon in a 3D Grid– A geological surface in the 3D Grid. The main
difference between a horizon and a surface in Petrel is that a horizon
uses a 3D rather than a 2D Grid. This means that it can have multiple z
values at a single XY value, whereas a surface can not, as a result,
reverse faults can be accounted for.
Intersection– A plane along which data can be displayed. These may
be planes in any direction, model grid lines, seismic lines, well paths or
intersection fences. Intersections can be displayed in 3D or in a 2D
intersection window ready for printing.
Kriging– Is a linear equation system where the variogram values are
known parameters, and the kriging weights are unknown parameters.
Model– A grid or group of grids based on the same fault structure and
boundaries. Each project can contain several models and each model
can contain several 3D grids.
Nodes– Points in the 3D grid where pillars are intersected by horizons.
Petrophysical Modeling – Interpolation or simulation of continuous
data e.g. porosity, permeability and saturation.
Probability – A measurement of the likelihood of an event (measured
in percent).
Property Models– Data on petrophysical properties held within each
cell of the 3D grid.

Property Modeling Introduction • 25

 Seed– Is a random number which organizes together with a formula
the visitation order in a 3D model.
Scale Up Well Logs– The process of sampling values from well logs
or well log attributes into the grid, ready for use as input to facies
modeling and petrophysical modeling.
Standard Deviation– This number is actually just the square root of
the variance, and is also used when describing how the distribution of
data points varies from the mean.
Stationarity– Is simply an assumption which is made regarding the
rules for behavior of the properties which we analyze, study, or model
with geostatistical tools.
Stochastic Simulation– Based on a given input data set and the final
result will be calculated on a probabilistic framework. The output
results are multiple realizations depending on the input data and a
random path.
Surfaces– A surface held in a 2D grid. Compare with the Horizon in a
3D grid. Not locked to the model (3D grid)
Template– An object describing the color table settings common to
groups of data. Petrel comes with several predefined templates,
including, depth and thickness color tables, property templates and
seismic color tables.
Toggle – Describes the action of switching objects and folders on or
off in the Petrel explorer panes for visualization purposes. E.g. toggle on
the Wells in the Input pane; this refers to the action of clicking the
check box in front of the Wells folder in the Input pane, when toggled
on a tick mark will be visual within the check box.
Transformation– Is a preparation of a real data set into an internal
data set which meets statistical requirements given by the chosen
algorithm.
Trend – Permanent or continuous change of the mean value of a
property in a 1D, 2D or 3D model.
Variance– A measurement of how spread out a distribution is. It is
computed as the average squared deviation of each number from its
mean (measured in square units of the collection).

26 • Introduction  Property Modeling

Variogram– Is a quantitative description of the variation in a property
as a function of separation distance between points, and is represented
graphically like plot of variance (Y) versus distance class (X).
Well Tops – Intersection points between well trajectories and
structural surfaces.
Zones– A zone is the volume between two horizons.
3D grid– A corner point 3D grid suitable for geological modeling and/or
flow simulation.

Acronyms used in this manual:

LMB – left mouse button, short for left-click
RMB – right mouse button, short for right-click
QC – quality check

Property Modeling Introduction • 27

Module 2 - Basic Statistics

This module covers basic statistical fundamentals, such as histograms,

cumulative distribution functions, crossplots, theoretical distributions
and variograms. These basic fundamentals form the basis of the
geostatistical tools and property modeling methods in Petrel.

Prerequisites
To successfully complete this module, the user must have knowledge of
the following:
• Petrel introduction

Learning Objectives
The purpose of this module is to give the participant a general
understanding of statistical concepts to become quickly involved with
the related tools in Petrel.
At the completion of this module, you will be able to:
• Understand the basic geostatistical concepts for Property
modeling
• Use the input distribution of a property to view in histograms,
prepare data, quality check and plot as cumulative frequency
• Understand the Normal distribution concept
• Use the Cumulative Distribution Function and transform data
into Normal Score for data analysis
• Use crossplots for correlation analysis between two variables
• Understand the principle of variogram calculation and
interpretation
• Calculate and model variograms in Petrel

Property Modeling Basic Statistics • 29

 Lesson

Basic Statistics
What is Geostatistics?

Geostatistics is a branch of applied statistics that places emphasis

on the geological context of the data and the spatial relationship
between the data

Geostatistical techniques are indispensable part of reservoir

management because quantitative numerical models are required for
planning the field/reservoir development to optimize time, resources
and economic gain

Geostatistics
What is Geostatistics?
In general, statistics is concerned with scientific methods for collecting,
organizing, summarizing, presenting and analyzing data, as well as
drawing valid conclusions and making reasonable decisions on the
basis of such analysis.
Up to 15-20 years ago statistics was only used in the Petroleum
Industry as a descriptive tool for measurement of mean (often only
arithmetic), max, min, etc. Recognizing that uncertainty matters and the
development of faster computers, has since led to a higher interest in
Geostatistics.
Geostatistics is built on statistic principles based on the concept of
the random variable in space. The use of Geostatistics began in the
gold mine industry in South Africa, when engineers Daniel Krig and
Daniel Wijs made an empirical sample and realized the increment of
variability. In 1962, Georges Matheron made the scientific framework
and developed the Kriging algorithm, still used today.

30 • Basic Statistics Property Modeling

Geostatistical techniques are an indispensable part of reservoir
management because quantitative numerical models are required for
planning and economic optimization. Monte Carlo Simulation (MCS) is
the key technology to quantify uncertainty.

Basic Statistics
Why use Geostatistics in Reservoir modeling?

Very few direct observations

Analysis of variables in space and
its correlation
Description of the reservoir
heterogeneity
Provide means for populating a 3D
model in a consistent and
reproducible way
Systematic way of describing and
handling reservoir uncertainty

Why use Geostatistics?

An important business need is to be able to make the best possible
decisions in the face of uncertainty due to the incomplete knowledge of
a dataset. One of the biggest uncertainties is the numerical description
of the subsurface.
When we compare how samples relate to each other, it is based on the
idea that samples close to each other have a better correlation.
Geostatistics provide us with a description of reservoir heterogeneity.
This gives a better estimation of reserves due to the calculation of
improved property models.
The ultimate goal is to optimize production by using the simulation of
the property models to decide on infill drilling wells, which will be used
for simulation predictions.

Property Modeling Basic Statistics • 31

 Basic Statistics
Definitions

Probability - a measurement of the likelihood of an

event. (Measured in percent).
Variance - a measurement of how different the
members of a collection are from each other. (Measured
in units of the collection).
Correlation - a way to measure whether two separate
collections are related. (Measured in percent).
Anisotropy - a way to measure whether variance within
a collection of data is determined by direction.
(Measured in azimuth and percent eccentricity).

Some examples
Probability: to see those locations in a saturation model where there
is a 70 percent probability that values will be greater than 0.6.
Variance: samples of porosity in a simple sandstone unit will show
much less variance than samples measured in a unit containing several
sands and a shale.
Correlation: values of seismic attributes sometimes show a strong
correlation with values of certain petrophysical properties.
Anisotropy: a porosity data set is anisotropic because measurements
in one direction vary more rapidly than in other directions.

32 • Basic Statistics Property Modeling

Basic Statistics
Definitions

Stationarity is simply an ASSUMPTION which is made

regarding the rules of behavior of the properties we
analyze, study, or model with geostatistical tools.
In practical terms, it means that the overall mean of a
property (e.g. average porosity) is constant and
differences from this mean are seen as local fluctuations.
The concept is applied in Geostatistical algorithms and is
linked to Standard Normal distributions (through Normal
score transformations)

Stationarity
The rule is that the property must behave consistently within the
volume chosen for analysis, study or modeling. If it does not, the
geostatistical tools which we use will not work properly. Stationarity
assumes that a property behaves the same way in all locations of the
chosen volume, for example, that the samples have no inherent trend. If
a trend exists, it must be removed (by transformation) before using
certain algorithms. Transformations are covered later in this module.

Property Modeling Basic Statistics • 33

 Basic Statistics
Histogram and Probability Distribution Function (PDF)

Classes

PDF

Property Value

Histogram is a graphical representation of the

frequency distribution of a selected variable

Histogram
A histogram is a graphical representation of the frequency distribution
of a selected variable and is given a data set of ‘n’ measurements
sorted in an increasing order. The interval between the smallest and the
largest value is divided into classes with equal value intervals. The
classes are displayed as a function of number of values belonging to
each class.
Each colored column represents a class (range of values) [X –axis].
The height of each column shows the number of points whose values
fall in the range of the class [Y-axis].
The overall shape of the histogram shows how the data may be
“grouped”.
Example: The group of points in the first class are smaller and appear to
be independent of the rest of the data, suggesting that they could be
excluded from the higher, more useful values of this attribute.
Normal (Gaussian) Probability Distribution Function (PDF)
Represents a symmetric, bell-shaped curve representing the Mean at
the peak and Variance as spread around the Mean.

34 • Basic Statistics Property Modeling

Definitions of the Histogram parameters
Class: A range of the property value. In the Distributions:
-Number of classes ~ Squareroot (n), where n = No of total
samples. For example, 100 data gives 10 classes.
-Number of classes can also be defined as ~ 1 + 3.32 lg*(n).
When you have a very high number of data.
Mode: The value that occurs the most frequently in a data set or a
probability distribution.
Median: The number separating the higher half of a sample, a
population, or a probability distribution from the lower half.
Typical Uses of Histograms:
Histograms are typically used to clean up log data and for quality
control after lumping, upscaling of well logs and modeling.

Basic Statistics
Cumulative Distribution Function (CDF)

CDF

0 Classes Property Value

The histogram classes are ranked in ascending order and

displayed as a cumulated fraction

Cumulative Histogram and CDF curves

In a cumulative histogram, the classes are repackaged in ascending
order by plotting the cumulative frequency along the Y-axis as a
summed fraction. A Cumulative Histogram is a valuable descriptive tool
and is used for inference.
Property Modeling Basic Statistics • 35
 From the Cumulative Histogram the Cumulative Distribution
Function (CDF) of normal distributed data is described analytically.
Through the CDF curve, data can be transformed to a different
distribution.
From PDF to CDF
By integrating the PDF over all values of x, the CDF of x is obtained.
Cumulative Frequency = Number of samples of each class /
Number of total samples
CumFreq (class 1) + CumFreq (class 2) + CumFreq (class 3) + ... +
CumFreq (class n) = 1

Basic Statistics
PDF and CDF in Petrel (Histogram)

Histogram & CDF in a Histogram window Histogram in object settings

1. Open a Histogram window. 1. Open the object Settings dialog.

2. Select the property to plot. 2. Go to the Histogram tab.
3. Select the Show cdf curve button. 3. Use filters and intervals/increments
4. Use filters if needed. as needed.

There are several different ways to visualize Histograms in Petrel:

Open a Histogram window from the Window menu and display the
variable (log or property). From here, the Cumulative Distribution
Function (CDF) can be calculated and shown. There are also different
filters available for visualization purposes (values, zones, facies, etc).
36 • Basic Statistics Property Modeling
In the Objects settings, under the Histogram tab, you can view the
histogram, and use different filters to display the values.
Other important information is shown in the object
Settings>Statistics tab, such as minimum and maximum value,
number of samples, etc.

Basic Statistics
Theoretical Distribution

Normal Distribution (bimodal) Gamma Distribution

10 1,00 14 1,00
12
8 0,80 0,80
10
Frequency

6 0,60 Frequency 8 0,60

4 0,40 6 0,40
4
2 0,20 0,20
2
0 0,00 0 0,00
5 6 7 8 9 10 11 12 13 14 15 16 More 5 10 15 20 25 30 35 40 45 50 55 60 More
Classes NV Classes Gam m a

Lognormal Distribution A histogram is a graphical help to find the

12 1,00
shape of the distribution (Normal, Lognormal
10 0,80 or Gamma)
Frequency

8
0,60
6
4
0,40 The Distributions have specific shapes and
2 0,20 parameters
0 0,00
5 7 9 11 13 15 17 19 21 23 25 27 More
The CDF curve shapes are dependant on the
shape of the histogram
Classes LNV

Theoretical distributions
The Distribution shape is an important aspect of the “description” of
a variable. It tells you the frequency of values from different ranges of
the variable. Typically, a researcher is interested in how well the
distribution can be approximated by the normal distribution. Simple
descriptive statistics can provide some information relevant to this
issue. For example, if the skewness (which measures the deviation of
the distribution from symmetry) is clearly different from 0, then that
distribution is asymmetrical, while normal distributions are perfectly
symmetrical.

Property Modeling Basic Statistics • 37

 Normal Distribution: The advantage is that if there is a large data
number, all data can be treated as normally distributed. The
disadvantage is that the distribution is sensitive to outliers, causing
skewness. Sandstone porosity is typically normal distributed.
Lognormal distribution: The property deviates from normal
distribution and has a skewed shape (Geometric mean).
Gamma distribution: Property deviates with extreme skewness; most
values are low (Harmonic mean). Typically, shale permeability has this
distribution.

We will focus on Normal Distributions because they are used as

standard normal in many geostatistical algorithms.
Normal Distribution of data means that most of the samples in a set
of data are close to the mean value, while relatively few samples tend
to one extreme or the other. The x-axis is the property value and the
y-axis is the number of data points for each value on the x-axis.
38 • Basic Statistics Property Modeling
Not all sets of data will have graphs that look perfect, some will have
relatively flat curves, others will be steep, sometimes the mean value
will lean a little bit to one side or the other. But all normally distributed
data will have something like the same “bell curve” shape.
The Probability of Normal Distribution (theoretical distributions,
endless data) is given by the formula p(x; μ, s), where:
p = probability
x= variable
μ = Mean
s = Standard deviation
e = is the base of the natural logarithm, sometimes called Euler’s e
(2.71...)
p = is the constant Pi (3.14...)
The equation after derivation will be for:
Normal Distribution: Artithmetic mean + Standard deviation
Lognormal Distribution: Geometric mean + Standard deviation
Gamma Distribution: Harmonic mean + Standard deviation

For a practical dataset (samples):

Mean is just the average of the property values (arithmetic mean).

, where:
x = Mean
= Variable
N = No of total samples
For example, what is the Mean of these numbers 3, 10, 5?
x = (3 + 10 + 5)/3 = 6
Variance and Standard Deviation are measures of how spread out a
distribution is, it means they are measures of variability.
The Variance is computed as the average squared deviation of each
number from its mean.
Property Modeling Basic Statistics • 39
 , where:
= Variance
= Variable
x = Mean
N = No of total samples
For example, for the numbers 3, 10, and 5, the Mean=6 and the
Variance is:

Standard Deviation measures data variability from the mean. It tells

you how tightly all the various examples are clustered around the mean
in a set of data. When the examples are tightly bunched together and
the bell-shaped curve is steep, the standard deviation is small. When
the examples are spread apart and the bell curve is relatively flat, then
you have a relatively large standard deviation.
How to calculate the Standard Deviation?
=sqrt(Variance), where:
S = Standard Deviation
= Variance
For example, for the numbers 3, 10, and 5, the Mean=6 and the
Variance=13

Then

40 • Basic Statistics Property Modeling

 Basic Statistics
Normal Score Transformation
 ( x−µ )2  x
2
− 
1 2s 2  1 − 2 
p ( x; µ , s ) = e − −− > p ( x;0,1) = e
s 2p 2p

Statistical Risk α (%) Factor in

confidence terms of
level standard
S=1 - α (%) deviation
68.3 31.7 1.000

90.0 10.0 1.645

95.0 5.0 1.960

95.5 4.5 2.000

99.0 1.0 2.576

99.7 0.3 3.000 Result: Transformation table in both

directions; data are transformed into
Standard Normal Distribution
The Normal Score Transformation is defined as the probability in
percent of how much of the distribution is within the area range when
Mean=0 and Standard Deviation=1. The data is now considered as
Standard Normal Distribution.
Example given from table when Mean=0 and Std Dev=1, 2 or 3
then 0 +/- StdDev for:
1Std Dev -1, 0, 1 Statistical confidence level is 68.3 %.
2Std Dev -2, 0, 2 Statistical confidence level is 95.5 %.
3Std Dev -3, 0, 3 Statistical confidence level is 99.7 %.

With a higher Standard Deviation more of the input data will be used
and the statistical confidence level will increase.

Property Modeling Basic Statistics • 41

 Basic Statistics
Normal Score Transformation

Normal Score Transformation

The histogram column (lower left plot) plots the values (porosity)
against the frequency (how often they appear within one bin/class).
When the histogram is plotted as a cumulative frequency (top left plot),
it gives a CDF curve.
When you perform Data analysis in Petrel, you have to do a Normal
Score Transformation of the data before you run the Sequential
Gaussian Simulation algorithm, since the algorithm assumes Normal
Scored data. Once the data is normal scored, the variograms can be set
up and the Simulation can be executed. In Petrel, after the execution,
the data will automatically be back-transformed from the normal
distribution to the original distribution.
Slide example:
The red arrow indicates that a value will be picked and simulated from
the normal distribution.
The green arrow indicates that after the geostatistical algorithm has
been run, the picked value is back-transformed to its original
distribution.
42 • Basic Statistics Property Modeling
 Basic Statistics
Crossplot and Correlation

Positive correlation Negative correlation

Crossplot
• Displays the values of two variables
measured at the same location
• Reveals the Degree of Correlation
(-1 to 1)

No correlation

Crossplot and Correlation coefficient

A Crossplot displays the values of two variables measured at the same
location and reveals the Degree of Correlation (-1 to 1) between both
variables by the shape of the data cloud.
Crossplot analysis and computation of Correlation Coefficient can
help identify useful secondary attributes, or rule those out that show
little or no correlation. The magnitude of the correlation coefficient is
the thing to look for.
When both variables increase, a Positive Correlation is shown, and
when one variable increases and the other decreases, a Negative
Correlation is shown. A strong negative correlation is acceptable, as
the relationship does not have to show a positive correlation to be
useful. In making the Crossplot, one of the variables is called Primary,
and the other is called Secondary. Typically, the secondary will be the
surrogate attribute.

Property Modeling Basic Statistics • 43

 One of the requirements for making a Crossplot is to have a reasonably
large set of data pairs where both target variables have been measured
at the same location. This ensures that the correlation between the two
variables will be considered significant. For example, a low correlation
coefficient with 100 pairs of points is better than a high correlation
coefficient with 10 pairs.

Basic Statistics
Correlation Analysis in Petrel (Function window)

Correlation setup:
1. Open a Function window
2. Select the properties to cross-plot;
Three properties may be plotted in
the same diagram (x, y and z-color)

A log scale on one or both

axis is possible

Select whether to view 3D

grid, upscaled cells
and/or raw well log data
If two properties are well correlated, then one may be
used as secondary input when modeling the other if this Linear regression and
has insufficient data (e.g. only a few wells) correlation coefficient

In Petrel, crossplots can be displayed in a Function Window and the

Correlation Coefficient can be calculated by making a Linear
Regression. After you have generated function and correlation
statistics, a new function will be stored in the Input pane. Open the
Settings to see the parameters. It can be edited by using the Select
and edit/add point icon from the Function bar.
This slide illustrates a Porosity vs Permeability crossplot showing a
strong correlation. The Permeability axis is shown as a logarithmic
scale (i.e. the strong correlation is a semi-log relationship). The
different colors represent different types of facies.
44 • Basic Statistics Property Modeling
 Basic Statistics
Generic Filters in Petrel (Histograms and Crossplot)
Different filters can be applied in Petrel, but Generic filters from the Histogram and Function
windows can be interactively applied to the 3D property models in 3D or plot windows.

How to generate a filter in Petrel:

1. Use the different filters from the
function bar for Histogram and
Function windows.
2. Create the filter by selecting an area in
the window (values of interest).
3. The new filter will be stored in the
Input pane > Filter folder > User

Generic Filters
1D filters – are based on a single attribute; can be defined in a
function window or in a histogram.
2D filters – are based on 2 attributes; can be defined in a function
window. The points to be filtered may be defined by a square or a
polygon drawn by the user.
Spatial filters – are based on their position in 3D space. These are
defined as a distance around a picked point, which may be picked in a
3D window or in the function window.
Logical filters – are based on a combination of any of the other filters.
To create or edit them, go to the Filter folder>right-click>Create/edit
logical filters. Filters may be constructed using any combination of
AND, OR and NOT commands with brackets to define the sequence in
which each filter should be applied.

Property Modeling Basic Statistics • 45

 Basic Statistics Part 1 – Exercises
This exercise has a pre-made Petrel project stored in the Projects folder:
Property Modeling [Link]. Double click to open the project.

Display and analyze a Histogram and a CDF curve

Exercise Steps
1. From the menu bar, open the Window menu, select New
histogram window.
2. Select the ‘Perm’ log from the Global well logs folder.
3. Click the Show cdf curve button .
4. Click on the Show viewport settings button to open
the Settings tab for the Histogram window and change the
data range to [Min: 0.1, Max: 2000]
5. Click the button and click Apply. See how the cdf curve
changes.
6. Deselect the Min. and Max. and click OK to close the Show
viewport settings button.
7. Select to use a part of the data in the histogram by clicking the
Select using 1D range on X axis button . On the
Histogram window, create a filter to avoid the Permeability
zero values, as shown in the figure below.

8. The selection is stored as a filter ‘Permeability 1’ in the Input
pane>Filter folder>User.

46 • Basic Statistics Property Modeling


Now, use the Generic Filter to make permanent changes to the well log.
9. Go to the Expl Wells sub-folder > right-click > Calculator
and use the Permeability_1 filter to remove the permeability
zero values. Enter the expression shown below. To insert the
filter, select it from the Filter folder and use the blue arrow in
the calculator.

When you use a

generic filter in the
calculator, it is
recommended to filter the
values that you want to
keep by setting the filter
equal to 1.

If a warning appears (cannot find the variable

Permeability_1), click OK and OK for all. It is because the filter
does not exist as a variable in the Well logs variable list.
10. In the Histogram window, deselect the Perm and select
Perm_temp from the Global well logs folder to visualize the
change.

Property Modeling Basic Statistics • 47
 11. The Perm_temp property will be continuously used in the
exercises for the next modules.
Create a Normal Distribution from a Histogram
Exercise Steps
1. Open a New histogram window. Under the main Wells
folder, expand the ‘Expl Wells’ sub-folder and from well
DW3>Well logs, select the PHI log.
2. Open the PHI>right-click>Settings>Statistics tab and look
for the Mean and Std. dev. values.
3. In the Histogram window>function bar select the Create
new distribution function button . A Create
distribution function window will open. Enter the Name of
distribution, select the option Normal distribution and enter
the values of the Mean, Std. and n.o. points (number of
points), as shown in the figure below.
4. Click on the Get from histogram button to specify the range
of the distribution and click Run to create the distribution.

The distribution
function can be interactively
edited by using the options
Select and edit/add points
and Select and edit line in
both windows (Histogram
and Settings>Function tab).
5. The new distribution function is stored in the Input pane.
Analyze it in the Histogram window or open PHI Normal
48 • Basic Statistics Property Modeling
 Distribution > Settings > Function tab, where the
Probability of a PHI Normal Distribution with Mean=0.10 and
Std. dev.=0.11 can be visualized.
The PHI Normal
Distribution can also be
automatically generated
using the PHI parameters by
selecting the Method Fit
normal distribution to
active histogram

Display a Crossplot and calculate the Correlation

Coefficient
Exercise Steps
1. Open a New function window from the Window menu.
2. Under the main Wells folder, expand ‘Expl Wells’ sub-folder
and from well DW4>Well logs select PHI on the X-axis and
Perm_temp on the Y-axis (in the previous exercise, this was
edited to cut away all zero values, and will create a nicer
correlation coefficient).
3. Use the Facies log as a third variable (Z) to color and to see
how the distribution of facies relates to the Permeability-
Porosity. To view the facies types, turn the Autolegend.
4. Display the Perm_temp (Y-axis) in logarithmic scale by
clicking the appropriate button in the function bar.
5. Calculate the correlation coefficient between the two logs
by clicking the Make linear function from crossplot button
in the function bar. The correlation coefficient is
Property Modeling Basic Statistics • 49
 displayed in the pop-up window. When clicking OK in this
window, the linear function is stored at the bottom of the Input
The Linear function
expression is stored in the pane.
Comments tab under the Info Optionally, the Function edit tools and Generic Filters can be used.
tab of the Settings of the
‘Perm_temp_vs_PHI’. 1. Change the visual settings of the correlation line in
Settings>Style tab of the ‘Perm_temp_vs_PHI’ function.
2. In the Function window, the new function can be edited by
using the Select and edit/add points and Select and
edit line .buttons. Open the ‘Perm_temp_vs_PHI’
Settings>Function tab and see the effect on the function line.
3. Now, filter part of the Well log data displayed in the crossplot
by clicking either the:
a. Select using 1D range on X axis button ,
b. Select using 1D range on Y axis button ,
c. Select using freehand draw button or
d. Select using 2D rectangle button .
The selection is stored as a filter in the Filter folder>User in
the Input pane.

In this case, Generic Filters can be used for Wells or Properties to

filter the data of interest and to calculate its correlation coefficient by
using the Make linear function from crossplot icon. Here, we will
use the filter, assuming it has filtered out the good reservoir part of the
data:
4. In the Function window by displaying one of the Generic Filters
that you have generated previously, calculate the correlation
50 • Basic Statistics Property Modeling
 coefficient using the Make linear function from crossplot
icon (restricted to one well at the time or a crossplot).
See an example in the figures below:

Comments
The Distribution Functions are useful in other processes, such as in
Facies modeling and Petrophysical modeling, to define the facies
probability distribution and properties distribution into the process
dialog respectively.
The Correlation Coefficient analysis is helpful to define the
correlation between properties by zone (defined with the Filters tools).
In Facies and Petrophysical modeling, the Correlation Coefficient can be
applied for secondary data during modeling.
Both processes are a useful quality control tool pre/post modeling.

Property Modeling Basic Statistics • 51

 Lesson

Basic Statistics
Variogram Concept

Variogram:
A quantitative description of the variation in a property as a function of
separation distance between data points
Based on the principle that two points close together are more likely
to have similar values than points far from each other

Two main aspects of a variogram:

1. How similar are two values right next to each other?
2. How far apart are two points before they bear no relation to each
other?

Variograms
The Variogram is used to model the way two values in space or time
are correlated. In general, two values in space that are close together
tend to be more similar than two values farther apart. Univariate
statistics cannot take this into account. Two distributions might have
the same mean and variance, but differ in the way they are correlated
with each other.
A variogram can be displayed as a plot of Variance (Y) versus
Distance Class (X). The variogram measures unique characteristics
about the variance of the data set, such as its Range, Sill and Nugget,
which will be defined in the next slides.

52 • Basic Statistics Property Modeling

 Basic Statistics
Variogram Parameters

Variance: A measure of how different

members of a collection are from eachother. Sill

Lag distance: Separation distance between

points.
Nugget
Sill: Variance at the point where the Range
summary plot flattens out to random Separation
similarity. 1 2 3 4 5 distance (lag)

Range: Correlation distance; distance The Variogram can be calculated in 3

beyond which data points no longer exhibit directions:
any statistical similarity.
• Horizontal Major
Nugget: Degree of dissimilarity at zero • Horizontal Minor
distance. • Vertical

Variogram Parameters Definitions

Variance: Average degree of difference between pairs of points. In the
Variogram plot, normally the Semi-variance is shown on the Y-axis.
Lag: Separation distance between data points or samples.
Sill: The semi-variance where the separation distance is greater than
the range. Describes the variation between two unrelated samples.
Transformed data should have a value of 1 and values much higher or
lower than this may indicate a spatial trend.
Range: Describes where the variogram model reaches the separation
distance where there is no longer any change in the degree of
correlation between pairs of data values. In other words, two data
points which are further apart than the range have only a random
relationship.
Nugget: The semi-variance where the separation distance is zero.
Describes the short scale variation in the data. Often, this is most
accurately identified from vertical data where the sampling interval is
usually much lower.
Property Modeling Basic Statistics • 53
 In summary, the closer two points are to each other, the more similar
they are expected to be (small variances for low separation distances).
The nugget represents small scale variation. This is often due to
measurement error, which should not be taken into account when
establishing the variogram model. Small scale variation which should
be taken into account when modeling is, for example, lamination, which
could cause rapid changes in porosity over short distances.

Basic Statistics
Variogram Calculation
Search radius
Calculation process and setup: Defines how far
1. A Search radius and Lag increment must away data will be
be defined => Decides the Number of lags collected
and consequently the Lag Distance Lag distance
2. All pairs of points in each Lag (bin) will be Defines max distance
compared pairs of data should be
3. For each Lag (with a given number of apart when compared
pairs), the average variation is calculated
(squared difference) (within each lag)
Semi variance

Variogram plot:
1. The Semi-variance vs. Lag distance is
plotted. These points (average variance per
lag) make up the Experimental Variogram
(black points) End of data
2. A Regression curve (grey line) is made range
based on all plotted points Lag distance
3. Fit a curve through the Experimental
Variogram to create a ’best fit’ Variogram Search readius
Model (blue line)
Lag distance

How is a Variogram computed?

• Define a Lag increment (or spacing between any two points in
the data), like, for example, 10 m.
• From the measurements, for all pairs of values separated by a
lag distance (for example, 10 m) - the difference is calculated
and then squared (seen as histogram bins in Petrel; equal to a
h-scatterplot).
54 • Basic Statistics Property Modeling
 • Sum up all the differences and divide by the number of pairs
(N). This gives you the value of the variogram for that particular
lag distance (one dot in the variogram). To get a
semivariogram, divide by 2N (see equation and example on the
next slide).
• The same is done for other lag distances like 20 m, 30 m, 40 m,
and so on.
• Plot out the variogram value versus the lag distance. You will
then get an Experimental Variogram (also called sample
variogram or simply the variogram).
• The result of a curve fitting (blue line) exercise based on the
experimental variogram points helps to explain the final
variogram shape, which is called a Variogram Model.
In summary, the points posted in the figure are referred to as an
Experimental or Sample Variogram. These are computed from the data
and each shown point represents a measure of average variation
(Variance) at a given separation distance (Lag).

Basic Statistics
Example of Experimental Variogram calculation

Semi-variance for 1 Lag distance Semi-variance for 2 lag distance

φ1 φ1
)
2

φ2
) ( φ2 - φ1 ) φ2 ( φ3 - φ1 )2
2

φ )
+ 2
φ3 ) (φ3 - φ 2 )
+
3 (φ 4 - φ 2 )
… + … +
… φi …
φi
φi +1 )
)
+ 2
( φi +1 - φi )
2
φi +1 (φi + 2 - φi )
… + φi + 2 +
… … …

N1 2 N2 2
1 1
g (h1) =
2 N1
∑ ((φ ) − (φ ))
i =1
i +1 i g (h 2 ) =
2N2
∑ ((φ ) − (φ ))
i =1
i +2 i

Property Modeling Basic Statistics • 55

 This slide illustrates a concrete example of how the experimental
variogram value is computed for a lag distance of one and two. 
The variance (gamma) for one and two lag distances (h1 and h2) is
equal to the Sum of the squared differences for all corresponding data
pairs divided by two times the number of data pairs (N).

Basic Statistics
Result of Experimental Variogram calculation
Nh 2
Semi-variogram 1
g =
can be calculated experimentally as: (h ) 2 N h ∑ ((Φ
i =1
(i +h ) ) − (Φ ))
i

EXERCISE
A WELL with a string of porosity values in depth steps of 1m: 3, 5, 7, 6, 4, 1, 1, 4.
Calculate the variogram values for lags 1, 2, 3, and 4 m respectively.
Plot the variogram. Is there a pattern?
Φ =3
Φ =5 g(h)
Φ =7
8
Sill
7 g(4)=7.124
Φ =6 6
g(3)=7.1
Φ =4 5

Φ =1 4 g(2)=5.75
Φ =1
3

2
g(1)=2.214
Φ =4 1

h
0
0 1 2 Range
3 4 5

In this example the semi variance value (gamma) for 1 lag distance
will be:
Gamma1 = 1 / (2* 7) * { (5-3)2 + (7-5)2 + … } = 1/14 * {4 + 4 + 1 + 4 + 9
+ 0 + 9} = 31/14 = 2.214
Variance value for 2 lag distance:
Gamma2 = 1 / (2*6) * { (7-3)2 + (6-5)2 + …} = 1/12 * {16 + 1 + 9 + 25 + 9
+ 9} = 69/12 = 5.75
Variance value for 3 lag distance:
Gamma3 = 1 / (2*5) * { (6-3)2 + (4-5)2 + …} = 1/10 * {9 + 1 + 36 + 25 +
0} = 71/10 = 7.1
56 • Basic Statistics Property Modeling
Variance value for 4 lag distance:
Gamma4 = 1 / (2*4) * { (4-3)2 + (1-5)2 + …} = 1/8 * {1 + 16 + 36 + 4} =
57/8 = 7.124

Basic Statistics
Variogram Model Types
Variance

Distance

Spherical: Good general algorithm

Exponential: Produces the most “noisy” result
Gaussian: Produces the smoothest result

Variogram Model Types

Variogram model is a continuous mathematical expression used to
describe the Experimental Variogram. The variogram model in Petrel
also contains information of anisotropy. When you want to fit the
Model Variogram curve to the Experimental Variogram (Sample
Variogram) you need to define the Model Type (shape of variogram
model curve).
There are 3 variogram model options in Petrel:
Spherical: This type is the simplest and has a linear behavior at
shorter distances with a sharp transition to a flat sill. Spherical
variograms are the most robust and stable in terms of the Kriging-
equation system to be solved. Influence of the data points are limited
by the range.
Exponential: This variogram has a steep behavior at shorter distances
with an asymptotic approach to the sill at longer distances.

Property Modeling Basic Statistics • 57

 Gaussian: This option is reserved for phenomena that show a high
degree of continuity at short distances and then begin to transition to
Exponential and
Gaussian variograms using a more of an exponential behavior at longer distances. Rarely used for
zero nugget effect can cause porosity, avoided for discrete data types, and used with caution for
problems solving the
kriging-equation system. A
permeability, when justified. Data points beyond the range have (some)
small nugget value will help influence on each grid node value.
in this case.
The Exponential and Gaussian models reach their sill asymptotically.
The ”Effective Range”, is the distance where the variogram reaches
95% of its maximum.

Basic Statistics Experimental Variogram

Applied Variogram Modeling
Variogram calculation process
Calculate the Experimental Variogram
Fit a Variogram Model to the Experimental Variogram
Variogram model types could be Spherical, Gaussian or
Exponential

Interpretative Process
Should take geological knowledge into account
Variogram Model
Vertical Variogram Model
Usually plenty of data and easily estimated

Horizontal Variogram Model

Can often not be calculated due to limited amount of data
Can be derived from correlated data source or taken
from analogous field / outcrop / geological knowledge

For carbonate Applied Variogram Modeling

reservoirs (with many
horizontal production wells) Well data is often far too sparse to facilitate variogram modeling in the
the situation is often vice horizontal direction. For example, if you have a single exploration well;
versa, and means a lot of
data in the horizontal this situation supports only vertical variogram analysis, no pairs are
direction and less data in the available in the horizontal plane. In this case, it is common to turn to a
vertical direction. Creating correlated secondary source of data. Given a reasonable correlation,
horizontal variograms will
then be easier than creating one can justify the horizontal analysis on the secondary data (e.g.
vertical variograms, but one variogram from seismic when modeling porosity). These data is used as
must be aware of scale
differences, cell size and
a proxy or substitute for the purpose of interpreting the direction and
support effect. major and minor range values.
58 • Basic Statistics Property Modeling
 Basic Statistics
Applied Variogram Modeling - Cyclicity

Variogram behavior example: Cyclic log porosity data due to varying facies vertically

Cyclicity = Hole effect

Search radius:
130 m

Search radius 130 m

Variogram Behavior - Cyclicity

Normally, a variogram shows larger variability with larger distances,
but sometimes, such as in fluvial systems, it can show Cyclicity.
The cyclic “humps” of the Experimental Variogram are called Hole
Effect and they yield a repetition of samples caused by geologic
processes.
A cyclic trend can be seen in the example zone porosity data. By looking
at all facies within a zone, such Hole Effects can be seen and the
search radius will limit how much data is compared. The amount of lags
are also important when searching for Cyclicity. It is more difficult to
see this effect when you are using many lags.

Property Modeling Basic Statistics • 59

 Basic Statistics
Applied Variogram Modeling - Trend
Variogram behavior example: Vertical Trend due to diagenetic effects, compaction etc.

A trend should be removed before

Variogram Modeling as it violates the
assumption of Stationarity Trend in data
Variogram modeling should be performed with
transformations as follows:
User: In the Data Analysis process, choose
between 1D, 2D or 3D Trend transformation
User: Verify the trend and correlation coefficient
Petrel: Model the residual by doing a function
trend There is no apparent Sill
Petrel: Add trend and residual to obtain
estimates
From CV Deutsch, 2002
Variogram Behavior - Trend
Another issue to search for within the data is known as a Trend. A
Trend is a systematic change of the mean value of the property in 1D,
2D or 3D. As an example, fining upward sequences will give variograms
without an apparent sill.
By applying a 1D, 2D or 3D Trend Transformation, a trend can be
detected in Petrel by the Correlation factor in the Data Analysis (seen
on later processes). A high Correlation factor related to the number of
pairs indicates a trend in the data set. It must be removed (practically
by applying trend function in the regression curve) to avoid a bias in the
output result. Without detrending, small values will be overestimated
and high values underestimated.

60 • Basic Statistics Property Modeling

 Basic Statistics
Why Variogram Modeling?

Requirement for geostatistical algorithms in Reservoir Modeling

Variograms are useful as Data Analysis tools

- Determine Layer thickness
- Determine directions/degree of Anisotropy
- Determine correlation/connectedness of facies data

Used as Quality Control to compare data before and after modeling

process

Why do we need Variograms?

1. Variograms are required for many geostatistical algorithms
To use certain geostatistical algorithms for modeling we must have a
Variogram, as the Variogram becomes the primary weight function
during modeling (see Kriging chapter).
As an example, if we choose a Kriging algorithm it needs a smooth,
continuous Variogram that fulfills specific mathematical properties. A
Variogram model that approximates the sample Variogram as closely as
possible fulfills these conditions.
2. To determine the natural heterogeneity of the data in the
vertical direction
The Range of the Vertical Variogram is a good indicator to define the
layering increment within a particular zone. We want a layer thickness
which will allow differences in facies to be seen (for more details, see
Module 7 - Quality control tools).
3. To determine if there is Anisotropy in the horizontal direction
During the creation of the horizontal variogram there are tools available
which allow us to establish if Anisotropy exists and then measure it
(see next slides).

Property Modeling Basic Statistics • 61

 4. To have another Quality Control measurement for comparison
before and after the Modeling process
As we analyze and model data, using the Variogram as the histogram is
an excellent tool for making sure that the characteristics of the data are
preserved after each modeling operation. If the proper algorithm was
used, the original well logs, upscaled well logs and 3D model
variograms should be similar. (for more details, see Module 7 - Scale up
well logs)

Basic Statistics
Anisotropy
Anisotropy is a characteristic of a data set, if there is a clear difference in how data
values change in a preferred direction.
If you suspect this kind of directional bias in your data set, incorporate that information
in the variogram to get a more accurate model.

The variability of
particle size across
the channels will be
much higher than
along the channels

Anisotropy is a way to measure whether variance within a collection

of data is determined by direction. It is measured in azimuth and
percent eccentricity.
Assuming Anisotropy:
How do you determine if directional bias exists in your data?
How do you measure the direction of bias in the data?
There are essentially two methods to determine anisotropy:
1. Trial and Error with directional Horizontal Variograms.
2. Using the Variogram Surface (map), if available.
62 • Basic Statistics Property Modeling
Basic Statistics
Variogram Maps and Sample Variograms in Petrel
In the object Settings > Variogram tab, you will find the available options for generating a
Horizontal variogram map and a Sample variogram for a property or correlated attribute.

Variogram Map Sample Variogram

Good for visualizing anisotropy Good for finding Major and minor
and its direction. Range horizontally.

In Petrel, Horizontal variogram maps and Sample variograms can

be made from the Settings dialog > Variogram tab of a property or
object attribute. The Variogram tab may be used in conjunction with
property filters to optimize the run time and avoid CPU intensive
processing.
There are many Variogram Types:
Classical Variogram – used as default in Petrel.
Pairwise Relative – each pair is normalized by the square average.
Logarithmic – logarithmic values are used instead of the original
values.
Semimadogram – uses the absolute difference instead of the squared
difference.
These all give different descriptions of spatial continuity. The objective More details about
different variogram types are
is still to help identify the range, yet each type affects the range in available in the Online Help
different ways. Manual.

Property Modeling Basic Statistics • 63

 The limitation is that the Variogram model should be based on the
Classical Variogram. The sill and nugget of all other variograms
cannot be used by Kriging and simulation. Only the range can be
derived from any type of variogram.

Basic Statistics
Directional Variogram Analysis (search cone)
Because of irregular spacing of input points, a Search area (defined by a Search cone) must
be defined in the search for points lying within the distance range given by the Lag.
Y axis

Angle=60o

X axis
Suggested Lag distance: Lateral = well spacing
Vertical = cell thickness

The Variogram ‘search cone’

Search radius, direction, angular tolerance, bandwidth, lag, and lag
tolerance define lag “bins”. Data pairs are identified based on a lag bin
methodology. All data pairs to the same base lag contribute to the
experimental variogram value for that respective lag distance.
Bandwidth: A distance cutoff used to prevent the lag bin search area
from becoming too wide at lag distances far from point of origin.
Angle Tolerance: Because it would be too restrictive to expect all
pairs in a given direction to lie along the exact line representing the
selected direction, Angle tolerance provides some leeway, identifying
data pairs which approximate a given direction without being too
restrictive.
64 • Basic Statistics Property Modeling
Lag tolerance: Distance +/- the lag spacing within which the data will
be considered a part of a given lag. Typically, this tolerance defaults to
½ of the lag spacing, ensuring that all data pairs within the maximum
lag distance and angle tolerance-bandwidth end up contributing to
some lag distance or another. As most analysis on upscaled well log
data has limited data to begin with, it is important not to lose any data
pairs.
Search distance: Is the maximum distance where you would like to
explore the variogram or the biggest lag. Normally, the extent of the
data or less.

Basic Statistics
Variogram Map – Theory
A Variogram Map is a way to present Variograms that have been computed in several
different directions over a data set (in Petrel: A point data set, surface or 3D property).
It produces a contour of the 2D variance surface (direction and extent of Anisotropy).
Y axis

Y Range
Number of
Y Lags

Number of X Lags
X Range

Note: The variogram map has its center at

coordinates (0,0) . It can, therefore, only
be displayed in a Map window in Petrel

X axis

Variogram map theory

A Variogram map is a way to present variograms that have been
computed in several different directions. This is done in an automated
fashion (that is, the various directions are computed for you in one
execution).
It produces a contour of the 2D semi-variance surface for
unambiguous detection of the direction and extent of anisotropy. If
Property Modeling Basic Statistics • 65
 anisotropy exists, you will see oval shaped contours whose highs or
lows will be in the center of the map. Think of the oval contours as sets
of small to large-sized ellipses whose major axis shows the major
direction of anisotropy.
The center of such a variogram map represents 0.0 lag distance. Out
from this center, the lag distances will show an increase in several
directions. In Petrel, variogram maps are displayed as a surface or
points. The grid geometry of such surfaces is in +/- lag space and is not
located in the project coordinate area. For this reason, variogram maps
do not have coordinates and can only be displayed in a Map window in
Petrel, and not along the structures in a traditional project base-map
display.

Basic Statistics
Variogram Map – Computation in Petrel

1. Select the Model Type. 4. Open a Map window and display the new
2. Define the parameters on XY range tab: Variogram map-
Number of lags and Search distance. 5. Use the Measure distance button to measure
3. Click Run. The result will be placed in the the Anisotropy orientation .
Input pane or 3D Grid > Variograms folder. 6. Read the values in the status bar.

4
1
5

Variogram maps in Petrel

To generate a Variogram map in Petrel, go to the Settings>Variogram
tab of a property or object, select the option Horizontal variogram
map and define parameters in the XY range tab (Number of lags in X
and Y, Search distance in X and Y) and Run the process.
66 • Basic Statistics Property Modeling
The Number of lags can generally be defined as follows: Search
distance divided by the average spacing of the wells (horizontally). The
Search distance can be defined as 70% of the model area to avoid
the edge effect (as a rule of thumb). However, this is just a starting
point, and if the resulting variogram map is not satisfactory, it will be
necessary to vary these parameters.
The Search distance is defined to calculate the variance between
pairs of points in both X and Y directions (E-W and N-S respectively). A
Variogram map actually represents the semi-variance in all directions,
but is just selected the maximum search distance for X and Y and the
process defines a search ellipse if they are not equal (a circle if they
are). If the result look different, it depends on how data is spatially
spread - usually the search distance is defined as at least half of the
field size in order to capture the variance of most of the pairs of points
on each direction. If the data for instance is aligned N-S, a higher Y
search distance than X can be used.
Select the new variogram map to be displayed in a Map window
(stored in the Input pane or 3D Grid/Variograms folder). To measure
the direction and extension of Anisotropy, use the Measure distance
button and position it in the center of the Variogram map. Follow the
major axis of the ellipses and then read the data on the information bar
at the bottom of the Petrel interface.
In the Variogram map, some holes can appear if there is no variance
calculated for a determined lag. It will often happen if there is not
enough data or if the spacing of the data is heterogeneous. It is normal,
but to avoid it, the ranges and number of lags must be adjusted to get The Variogram map
pairs of points in each lag. is placed in the Input pane if
it was generated from a
The following slide shows a variogram map displayed with contour line general object (e.g., points),
and color fill contours in a map window. Notice the obvious anisotropy and it is placed in the 3D
Grid/Variograms folder if it
(that is, the highly elliptical contour shape). This makes it quite easy to was generated from a
use the measuring tool to record the major direction for your variogram property.
model. The minor direction is 90 degrees to the major direction.

Property Modeling Basic Statistics • 67

 Basic Statistics
Variogram Map – Anisotropy

R minor R major

Variogram Map: Sample Variograms :

Arrows show the Major and Major and Minor ranges are
Minor direction of Anisotropy determined on the x-axis

Geometric Anisotropy
If a variable exhibits different ranges in the major and minor directions,
then there is Geometric Anisotropy. For example, in a shoreface
deposit, permeability might have a larger range along the shoreline
compared to the range perpendicular to the shoreline.
Geometrical Anisotropy is symmetric in both directions. Theoretically,
the Sill is the same and the Range is different.
Zonal Anisotropy
If the variable exhibits different sills in different directions, then there
is a Zonal Anisotropy. For example, a variogram in a vertical wellbore
typically shows a bigger sill than a variogram in the horizontal direction.
Zonal Anisotropy, one-directional, is handled like isotropic data because
the range of the major and minor direction is the same. Theoretically,
the Sill is different and the Range is the same.
This represents a limited case of Geometric Anisotropy. The range of
correlation in one direction will exceed the field size. This yields a
variogram that does not appear to reach the Sill/Variance.

68 • Basic Statistics Property Modeling

 Basic Statistics
Sample Variogram – Theory

Important model parameters:

Model type
Nugget
Range
Anisotropy (Azimuth given from
Variogram map)

These parameters must be the same

for the Experimental Variograms:
• Nugget
• Sill
• Variogram Model type

Note: Sill has no influence on

Kriging/Simulation estimation result

Sample variogram theory

A Sample variogram is calculated for a sample data set using a
direction and separation distance. Upon computing a Sample variogram
it is important to use a densely sampled and correlated property/object.
Once the major and minor directions have been determined from the
Variogram map, it is easy to produce Sample variograms in the two
directions. The calculated variogram model can be displayed in a
Function window. The major and minor ranges can be modeled in this
Function window as well. This approach avoids time-consuming
variogram computation on large property grids and can be used to
uncover potential anisotropy, along with variogram range in the major
and minor direction.

Property Modeling Basic Statistics • 69

 Basic Statistics
Sample Variogram – Computation in Petrel
1. Select the Model Type (Classical) and 3. Define the parameters on the XY range tab
Sample variogram. (No of lags and Search distance)
2. Define the parameters on the Orientation tab 4. Click Run to get the Major variogram
(azimuth given from variogram map). range. Repeat with 90 degree orientation to
get the Minor variogram range.

Sample variograms in Petrel

To generate a Sample variogram in Petrel, go to the Settings
dialog>Variogram tab of a property or object, select the option
Sample variogram and define parameters in the Orientation and XY
range tabs and Run the process.
The Orientation can be defined using the anisotropy direction value
from the Variogram map. To generate two orientations (major and minor
direction), the process must be Run twice in the respective directions.
The Number of lags can generally be defined as follows: search
vertical distance divided by average cell thickness. The search
vertical distance can be defined as 70% of the zone thickness to
avoid the edge effect (rule of thumb).

70 • Basic Statistics Property Modeling

 Basic Statistics
Variogram modeling – Computation in Petrel

5. Open a Function window and display the new Sample

variogram
6. Select Make variogram for sample variogram icon 7
7. Split into both major and minor variogram ranges using the Select
and edit/add point icon
8. Open the Variogram Settings to see the Variogram model
parameters (Model type, Sill, Nugget, Orientation and Ranges)

5
6
Sill

Nugget Range 8

Variograms modeling in Petrel

To model the Sample variogram, display it in a Function window.
To fit a Variogram model (for example, a Spherical model) to the Sample
variogram, use the Make variogram for sample variogram button.
Afterwards, a model variogram must be matched with the Sample
variogram to create a best fit model variogram curve.
This eliminates the need to experiment with the direction for horizontal
variogram analysis altogether. Simply transfer the direction, along with
The Sample
the major and minor range, to the respective primary property in the variogram is placed in the
Data analysis process and model the vertical variogram to obtain the Input pane, if it was
generated from a general
nugget, sill, and vertical range values. When saved, this variogram will object, or 3D
contain variogram parameters that reflect the property data in the Grid>Variograms folder if
vertical direction, but have horizontal parameters derived from the it was generated from a
property
correlated second data via the Variogram map and Sample variograms.

Property Modeling Basic Statistics • 71

 Basic Statistics
Variogram Modeling in Petrel
Settings/Variogram tab In Data Analysis process Property Modeling process

Can calculate variograms for all 3 directions,

on up-scaled well logs, raw well logs or 3D
Sample or Horizontal
property data Set up Range, Nugget and
Variogram to find anisotropy
Can easily see effect of search cone settings Azimuth directly into the
process dialog

There are essentially three options for defining the variogram settings
in Petrel:
1. You can do a horizontal variogram analysis from the object’s
Settings dialog (from the Variogram tab). Here, you can
compute a variogram map and compute sample variograms for
the two horizontal directions.
2. A more robust way of creating a variogram is to use the Data
analysis process in Petrel. This requires that the property
being analyzed must exist as a property under the active 3D
grid. You may do the analysis on raw well logs data, on
upscaled well log data or on the entire 3D property. You should
specify the transformations of the property before doing the
variogram analysis.
3. If, for some reason, you do not want to compute the variogram
(for example, if you just want to run quickly through your model
to see the effect of different settings, or if you are not able to
compute a proper variogram), then it is possible to simply enter
the variogram values (range, nugget, azimuth) directly into the
Facies modeling or Petrophysical modeling processes.
There are some default variogram settings already specified –
72 • Basic Statistics Property Modeling
 the only reason for this is that these variogram settings must
be specified to run these processes.

NOTE: You should never use the default settings unless you have
verified that they are appropriate for your field!

Basic Statistics Part 2 – Exercises

This exercise has a pre-made Petrel project stored in Projects folder:
Property Modeling [Link]. If you did the last exercise, please use
the same project.
Variograms are used as a method for describing spatial variation. It is
based on the principle that closely spaced samples are likely to have a
greater correlation than those located far from one another, and that
beyond a certain point (range), a minimum correlation is reached and
the distance is no longer important.
Of course, this spatial correlation can be anisotropic and several
variograms orientated in different directions may be required to
describe the variation in a property.

Variogram map calculation of a point data set

By generating a variogram from input data, it is then possible to use the
variogram when modeling properties thus preserving the observed
spatial variation in the final model.
In this simple exercise, you are given a point data set with two
attributes; depth and Acoustic Impedance (AI).
Exercise Steps
1. In the Input pane, go to the Variogram data set folder and
display the point data set Ac Impedance in a 2D window.
2. In the Settings>Variogram tab, select Horizontal
variogram map and change the settings under the XY range
tab according to the figure below. The Search distance set in
the variogram settings defines the new Variogram map
extension.

Property Modeling Basic Statistics • 73

 3. Click the Run button. The calculated variogram map will be
stored at the bottom of the Input pane.
4. Open a Map window and display the Variogram map. It is a
contour map (2D plot) of the sample variogram surface.
5. Click the View all in viewport button to see the
complete Variogram map. Determine the anisotropy direction
to be about 130 degrees by using the Measure distance
tool, positioning it on the map center and following the ellipse
major axis. The values will appear in the status bar at the
bottom of the window.
6. Keep the Variogram dialog open for the next exercise.

74 • Basic Statistics Property Modeling

Sample variogram calculation of point data set
Exercise Steps
1. Continue in the Variogram tab for the Ac Impedance point
data.
2. Generate a Sample variogram and select the parameters as
shown below in the picture. Use 20 lags and a horizontal
search radius of 26000 m. In the Orientation tab, enter -39
degrees of orientation (it is the equivalent to the orientation
measured in the Variogram map). Click Run. The Sample
variogram can be found at the bottom of the Input pane.

3. Display the Sample variogram in a Function window. Under

Settings>Style tab of the Sample variogram, you can change
the display style for Points and Line.
4. As a test, in the Settings of the data, set the Ac Impedance,
go to the Variogram tab and change the parameters as shown
in the table below and inspect the various results:

Property Modeling Basic Statistics • 75

 Orientation XY Range XY Range Variogram Type
No of Lags Horizontal Search Radius
-39 10 500 Classical
-39 10 30000 Classical
-39 100 30000 Classical
-39 30 30000 Classical

5. Calculate a new Sample variogram with an Orientation of

Notice the option
with a Horizontal search 51 degrees. Make sure that the check box Overwrite last is
radius of 500; a Sample NOT selected.
variogram cannot be 6. Display the different variograms in a Function window and
generated because the
process does not find any notice the differences between variograms calculated with
data in that small search using parameters.
radius.

Define a variogram model

The variogram model is a mathematical model used to describe the
sample variogram.
Exercise Steps
1. In the Function window, display the first calculated Sample
variogram of the major anisotropy axis (-39 degrees) that you
created in the last exercise.
2. Click on the Make variogram for sample variogram
button. A Variogram model will be displayed, and the model
is also stored at the bottom of the Input pane.
3. Display the second Sample variogram as well (51 degrees).
4. Now define the Variogram model range and nugget for both

The Variogram model displayed sample variograms interactively using the

type, the sill, and the nugget
must be the same for both button or the button. Notice, while editing the ranges
variograms. However, the Sill interactively, one point will correspond to the major range and
is of no importance for
kriging/simulation.
the other point to the minor range.
5. Right-click on the stored Variogram model and select
Settings. The parameters of the Variogram model for the
major and the minor anisotropy axis are stored there.

76 • Basic Statistics Property Modeling

Property Modeling Basic Statistics • 77
 Summary
In this module, you have become familiar with some statistical terms
which are important for Geostatistics. In addition to the basic terms,
the importance of distribution functions were highlighted. Graphical
tools, such as histograms and crossplots between two variables, are
essential to characterize a dataset.
The Variogram modeling process is a pre-step in geostatistical reservoir
modeling, and is separated into two important steps. First, the
calculation of an experimental variogram based on the input data, and
second, modeling in an effort to fit a theoretical function which fulfills
the requirements of the geostatistical algorithms. Maps are important
qualitative indicators of anisotropic behavior in the reservoir, whereas
sample variograms quantify that information. Variograms can also be
used for geological interpretation purposes.
All of the functionality will be used for Facies and Petrophysical
Modeling.

78 • Basic Statistics Property Modeling

Module 3 - Kriging

This module covers the basic concepts of the estimation technique

called Kriging and how to use it in Petrel. To test the Kriging
functionality and different parameters, we will focus on how to create
2D grids (surfaces in Petrel). This makes it easy to see the impact of
changing the input data and the variogram model settings.

Prerequisites
To successfully complete this module, the user must have knowledge of
the following:
• Familiarity with Geostatistical fundamentals
• Petrel introduction

Learning Objectives
The purpose of this topic is to gain an understanding of the Kriging
principle and the impact of the parameters in the result for 2D and 3D
property grids in Petrel.
This module covers:
• The principle of Kriging
• Types of Kriging
• Influence of variogram model parameters on Kriging results

Property Modeling Kriging • 79

 Lesson

Kriging
Concept
Kriging is an Estimation Technique/mapping method based on
fundamental statistical properties of the data (Mean and Variance).
Assumption of data set:
Stationarity
– Mean does not change laterally
– No trends
Inputs:
Data set
– Normally well data
Variogram and CDF curve
Use and Limitations:
Trend search and visualization of trends
Smooth and locally accurate results
Does not handle extreme values very well

Kriging
Kriging is an Estimation Technique solving a Linear Equation
System where the Variogram values are known parameters (Mean
and Variance) and the Weights are unknown parameters.
Kriging does assumptions of the data set, such as stationarity in
regards to the rules for behavior of the properties which we analyze,
study, or model with geostatistical tools. This means that the property
must behave consistently within the volume chosen for analysis, study
or modeling. If it does not, the geostatistical tools which we use will
not work properly. Stationarity assumes that a property behaves the
same way in all locations of the chosen volume; that is, the samples
have no inherent trend. If a trend exists, it must be removed before
using certain algorithms like Sequential Gaussian Simulation.
Kriging can be used for creating either a 2D surface or a 3D property
model in Petrel. Normally, input to the algorithm are well data as point
80 • Kriging Property Modeling
data and a variogram. The variogram defines how to use the well data
for populating values in the grid cells between the wells. The principal
idea is that the changes are interpreted local fluctuations around a
constant mean value.

Kriging
Estimation

Example:
On a 2D surface or a 3D grid the only known variable (e.g. Porosity)
is at the well location. All other grid cell values must be estimated:
n
z ( x0 ) = ∑ li z ( xi )
i

Unknown values z(x0) are estimated as a weighted sum of the known

values
Kriging Weights λi are calculated from Variogram model
Known values z(xi) are for example existing well data

In general, Kriging is a weighted moving average algorithm. The key

point for Kriging is how to calculate the Weights (li ).
The equation represents the Ordinary Kriging Algorithm. It combines
known values, after applying a weighting factor, so that an estimate
can be made at an unknown location:
• Known values z(xi) are values in the up scaled cells.
• Unknown value at position x0 is calculated from a linear
combination of surrounding data points.
• Kriging weights (li ) are calculated from a variogram model.
Generally weights decrease as you approach the range of the
variogram. Kriging uses the variogram to understand the
variability of the data over a distance.
• If position x0 is outside the Range of the data points, the value
at x0 will get the Mean value derived from the data, however,
Property Modeling Kriging • 81
 this does not always apply to Gaussian and Exponential
variogram model types.
Factors that can be considered in assigning the weights are the spatial
correlation between unknown and data, redundancy between data
values, closeness to the location that is being estimated, anisotropic
continuity and magnitude of continuity/variability.

Kriging
Weighted linear estimates
Kriging uses weighted linear estimates; i.e. combining known values
to estimate an unknown value Z at a location Xo
Influence of variogram
range on weights
n
X0 ? z ( x0 ) = ∑ li z ( xi )
i

Direction of major continuity

Weighting factor decided by Variogram:
Data with known values, Z(Xi) • How close to location?
• Preferred direction (anisotropy)
Value to be estimated, Z(X0)

Z (Xo) is not known, but we can still calculate the error variance because we know
the statistic parameters; mean value, variance value and the variogram model

Principle one: A datum close in “geological distance” to the unknown

should get a large weight.
Principle two: Data close together are redundant and should “share”
their weight.
Variance can be calculated internally during Kriging. The variance is
the square of the standard deviation at any point, and it can be exposed
as a property in Petrophysical modeling.
The variance property is a factor of the input data and the chosen
82 • Kriging Property Modeling
variogram, and describes some of the spatial variation in uncertainty in
the model. Areas close to the input data will have a low variance (that
is, are predicted with high confidence, low uncertainty), while areas
away from the input data will have high variance (that is, are predicted
with low confidence, high uncertainty).
A useful calculation of variance requires the correct sill in the data,
however this can lead to usability issues where the sill is incorrect, and
it is difficult to identify why the result is poor. Instead, the sill is fixed at
1 and the calculations are done internally to ensure that the correct
variance is reported based on the true sill of the data. If the sill needs
to be changed, then using a scale shift in the Data analysis process
can do it.

Kriging
Error Variance

How good is an estimate? This must be measured to choose the ’best’

estimate (Best Linear Unbiased Estimate – BLUE). The true value is
still not known exactly, so the Error variance can be calculated

The estimate that minimizes the error Minimize Error variance=Var[(Z * (u) − Z(u))2 ]
variance by construction is called Kriging
Kriging uses the variogram to understand
the variability of the data over a distance
This knowledge allows Kriging to calculate
weights that minimize the error variance
The ”Best” estimate in Kriging is a least
square criterion (where: sum of squared
residuals has its least value)

Variance
A Kriging variance output is given as a Probability distribution that
varies around the input data. The Kriging best local estimate gives the
highest probability.

Property Modeling Kriging • 83

 In Petrel Petrophysical modeling, this is the process of making a
variance output property:
1. Set up the parameters under Kriging interpolation / Kriging
/ Kriging by Gslib algorithms.
2. Select Create variance output in the Variogram tab. Click
the Create button first to create a new property.
3. Click Normalize to get a variance between 0 and 1.

Kriging
Types

Simple Kriging – mean value is known

n n
z ( x0 ) = ∑ li z ( xi ) + [1 − ∑ li ]m
i i

The sum of the weights li may be LESS than 1

The smaller the weights the bigger the influence of the Mean (m) on the calculated
value Z at position x0
A Global Mean value is used by the Kriging algorithm when calculating the weights.
The Mean value is assumed known
Default Kriging algorithm in Petrel

Kriging types
For Simple Kriging:
• Z(Xo) = Value at location Xo
• The sum of the weights should be <1. If the weights (lambda)
approach zero, the mean gets more and more influence.
• m = mean, it is an assumed constant known value and it is a
Global mean (more stable than Ordinary Kriging)
• Good for dense data.

84 • Kriging Property Modeling

 Kriging
Types

Ordinary Kriging – local estimation of mean

n
z ( x 0 ) = ∑ li z ( x i )
i

The sum of the weights li is ALWAYS 1

A Local Mean value is used by the Kriging algorithm when calculating the weights.
The Mean value is assumed constant but unknown

For Ordinary Kriging:

• It re-estimates the mean at each location, therefore, it is a
Local mean (constant but unknown).
• It is more unstable than Simple Kriging, but can be useful
when there are many wells.
• Can be suitable for data points that show a trend.
In Petrel, you have an option to use Ordinary Kriging instead of Simple
Kriging in the Petrophysical modeling process (Expert tab).

Property Modeling Kriging • 85

 How does kriging work in reality?
Kriging can be defined as an estimation technique based on the
interaction between the Variogram parameters and the local
neighborhood data. Starting from the Variogram analysis and the
existing data, Kriging combines the information in the following way:
The distance to the point to be estimated defines the data
configuration. In this example, a Variogram model has been set up and
the model has a known Variogram range (see figure above). The point
x0 should be calculated using the information from the points x1, x2 and
x3. The distances from point x1, x2 and x3 to x0 are in red. The
distances between the points x1, x2 and x3 are in blue. The variogram
was determined as shown in the figure and every distance is
representing a Variogram value.
For this procedure, we will use the Variogram spherical model to
calculate the variance for every distance; the nugget, sill and range
parameters will be used.

86 • Kriging Property Modeling

For example:
g(x0,x1)=g(40)=0+{1(1.5*(40/60)) – (0.5*(40/60)3)}=0.85
g(x1,x2)=g(60)=0+{1(1.5*(60/60)) – (0.5*(60/60)3)}=1
The Kriging equation matrix will be set up as shown in the next slide.

Kriging
From a Variogram Model to Final Estimation – 2

A linear equation system must be solved to find the value at xo.

A matrix with Variance between neighbour data and Variance from
neighbour data to estimated point is set up with Kriging weights
n
Linear Equation System to be solved: z ( x 0 ) = ∑ li z ( x i )
g ( x11 ) g ( x12 ) ... g ( x1n ) 1   l1  g ( x10 )  i
g ( x ) g ( x ) ... g ( x2 n ) 1  l2  g ( x20 )  Unknown weights
 21 22

 ... ... ... ... ... *  ...  =  ...  X2

      (0.22)
g ( xn1 ) g ( xn 2 ) ... g ( xnn ) 1  ln  g ( xn 0 )
 1 1 1 1 0   µ   1  60
30
X0 (?)
42
X1 40
l1 = 0.185 l2 = 0.291 l3 = 0.524 µ = 0.0425 (0.25) 20

55
Z(xo) = 0.185 * 0.25 + 0.291 * 0.22 + 0.524 * 0.30 = 0.267 X3
(0.30)

The linear equation system

To figure out the property value at location x0, we need to solve the
Linear Equation System. Where the gamma values are the
Variance between the neighborhood data and from the neighborhood
points to the point to be estimated, and where the lambda is the
Kriging weights.
The first matrix represents the Variance (variogram values) for the
distances between the neighborhood data (blue in the image). The
second matrix represents the Kriging weights and the third matrix is the
Property Modeling Kriging • 87
 Variance (variogram values) representing the distances from the
neighborhood points to the point to be estimated (red in the image).
Variance values (gamma) have to be found and multiplied with the
unknown weight (lambda).
The final result, an estimated value for Z (Xo):
A µ (Mju, La Grange multiplicator) is added to solve the equation
system. Following the example, the matrix with calculated gamma
values looks like this:
 0.000 1.000 0.990 1   l1   0.852 
   
 1.000 0.000 0.878 1  l2   0.688 
 0.990 0.878 0.000 * =
1  l3   0.481 
     
 1 1 1 0   µ   1 

This will be turned into a linear equation system:
0.000 * l1 + 1.000 * l2 + 0.990 * l3 + µ = 0.852
1.000 * l1 + 0.000 * l2 + 0.878 * l3 + µ = 0.688
0.990 * l1 + 0.878 * l2 + 0.000 * l3 + µ = 0.481
l1 + l2 + l3 = 1
This equation system will now be solved (usually) by a Gaussian
algorithm. The constraint in the Ordinary Kriging case requires the sum
of the weights to be equal to 1 (line 4). Kriging works like an Inverse
Distance Weighting procedure, but now the inverse Euclidean distances
will be replaced by the spatial equivalent – the Kriging weights .
Therefore, it is necessary to solve the Kriging equation system to get
the Kriging weights. In the example, the result would be:

The parameter µ helps to solve the equation system but is not

necessary for the final calculation of the estimation value Z(x0):
n
z ( x 0 ) = ∑ li z ( x i )
i

Finally, the value at the position x0 can be calculated:

Z(x0) = 0.185*0.25 + 0.291*0.22 + 0.524*0.3 = 0.267
88 • Kriging Property Modeling
 Kriging
Influence of the Variogram model parameters

Variogram Model Type (Range: 10000m/5000m)

Exponential model Spherical model Gaussian model

Variogram model type

In general, the Exponential and Spherical models give more similar Some Kriging
results (lower variability, more homogeneity in a short distances) than algorithms are slow compared
to other algorithms, but the
the Gaussian model. The Gaussian model gives a smoother result than new ‘Kriging’ method in Petrel
both Exponential and Spherical models do. is improved using
Parallelization

Property Modeling Kriging • 89

 Kriging
Influence of the Variogram model parameters

Variogram Range (spherical variogram)

Range: 1000m Range: 10000m

Variogram range
Hint/recommendation: The range should always be larger than the
average spacing of data points or at least larger than the data points
minimum distance.
In the example above, a large range results in higher connectivity and in
the opposite case a smaller range will result in lower connectivity and a
‘bulls eye’ effect.

90 • Kriging Property Modeling

Kriging
Influence of the Variogram model parameters

Azimuth (spherical variogram)

Bad Good

Anisotropic: +45 degree Anisotropic: -45 degree

Range: 10000m / 5000m Range: 10000m / 5000m

Variogram azimuth
If the Variogram search direction is oriented according to geological
know-how or a conceptual model, the result will be more geologically
meaningful. If not, the results of the existing data and a wrong
Variogram may contradict the geological understanding of the area.

Property Modeling Kriging • 91

 Kriging
Influence of the Variogram model parameters

Nugget (spherical variogram)

Anisotropic: -45 degree Anisotropic: -45 degree

Range: 10000m / 5000m Range: 10000m / 5000m
Nugget: 0.99 Nugget: 0.1

Variogram nugget
The example to the
left shows a nugget of 0.99,
Using a 100% nugget model (nugget = close to 1) will create maps of
meaning that kriging the properties that look continuous, except at the data locations, where
value at any points get the discontinuous jumps to where the actual data values occur (at the well
value of the global mean
(Simple Kriging) or a local positions). This happens because kriging honors the input data points.
mean (Ordinary Kriging).
The continuous appearance is due to the fact that any kriged point will
get the global mean (Simple Kriging) of the data when using a nugget
of 0.99.
If you have a small nugget effect (nugget = close to 0), the result will
have higher variability or heterogeneity.

92 • Kriging Property Modeling

 Kriging
Kriging algorithms available in Petrel

Kriging interpolation
It only considers data within the variogram range.
Kriging by Gsilb
It has Collocated co-kriging and added options in
Expert settings, but its performance is slower compared to Kriging.
Kriging
It is able to krige much larger numbers of points through a combination of
multithreading and smart neighborhood searching. It has parallelization, fast
Collocated co-kriging and additional Expert settings.

Preferred method

Kriging Interpolation: Searches data in XYZ space and populates data

in IJK space. It only considers data within the variogram range, which
may lead to strange effects in areas with no data when trends have not
been removed correctly.
Kriging by Gslib: Works in IJK space, has Collocated co-kriging and
added expert settings. Performance is slower compared to ‘Kriging’.
Kriging: Introduced in Petrel 2008.1. It is able to krige much larger
numbers of points through a combination of multithreading and smart
neighborhood searching. In Petrel 2009 and 2010, it has considerable
improvements through parallelization, fast Collocated co-kriging and
added expert settings. It works in XYZ and IJK space. It is the best in
terms of time performance.
For Kriging and Kriging by Gslib algorithms the Simple and Ordinary
Kriging types are available in the Petrophysical modeling process.
For all the versions of Kriging the use of an external Trend is available
in the Make/edit surface and Petrophysical modeling processes.
For more information about Kriging methods see the Petrophysical
modeling module: “Kriging algorithm”.
Property Modeling Kriging • 93
 Kriging – Exercises
Influence of the variogram model parameters on Kriging
results
The Kriging algorithm uses a variogram to express the spatial variability
of the input data. The user can define the model type of function for the
variogram (Exponential, Spherical or Gaussian) as well as the Range,
Orientation and Nugget.
This exercise has a pre-made Petrel project stored in Projects folder:
Property Modeling [Link]. If you did the last exercise, please use
the same project.
Exercise Steps
1. Display the data set Ac Impedance in the 2D window. You
will find it in the Variogram data set folder in the Input pane.
2. Open the Make/edit surface process under Utilities in the
Processes pane.
3. In the Main input, drop in the Ac Impedance point data,
select AI as attribute and select the check box Name; enter
“AI Kriging”.
4. Select the Geometry tab and select the User defined option,
highlight the Ac Impedance data and click the Get limits
from selected button.
5. Set the Grid increments X and Y to 200.

94 • Kriging Property Modeling

 6. Go to the Algorithm tab and select Kriging from the drop-
down menu. Click Apply.

Property Modeling Kriging • 95

 7. Now, calculate different surfaces using the variogram
parameters from the table below:

Range Major Range Minor Azimuth Nugget Model type

Dir Dir
500 500 0 0.1 Spherical
10000 10000 0 0.1 Spherical
10000 10000 0 0.9 Spherical
13000 7000 -40 0.1 Spherical

8. Display the results in a 2D or Map window. When the

surface is recalculated with new parameter settings, the old
result will be overwritten and the display automatically
updated.
Hint: When recalculating the surface to get a different output,
the Result surface must be removed by selecting it, deleting
it and it giving it a different Name. A pop-up window ask to
reset all the settings, select No to keep the old parameters and
only apply the necessary changes.
9. Change the model type from Spherical to Gaussian and
If the Nugget seems Exponential. Compare the results of all the models.
wrong, it is because the 10. Continue in the Make/edit surfaces process to calculate a
Variogram is scaled to a Sill
of 1 in the Make/edit
surface using the variogram model that you determined
surface, while it may be previously in the Sample variogram exercises (Module 2 Basic
different in the created Statistics Part 2 - Exercises).
Variogram model. Go to the
Variogram model settings 11. Leave the previous settings, but remove the Result surface
and toggle ‘Force sill to be output and give a new Name (“AI Kriging - Model Variogram”).
equal to 1.0’. 12. Select the Variogram model from the Input pane (“Sample
var. from Ac Impedance (-39 deg)”).
13. Go to the Algorithm tab. In the Variogram sub-tab, click the
button to get the parameters from the variogram
model.
14. Click Apply/OK to calculate the new surface and display it to
compare with the previous results.

96 • Kriging Property Modeling

Summary
In this module, you have learned that Kriging is used as basis for many
algorithms in reservoir modeling. It is an estimation method and will
generally create a smooth result. In general, Kriging will estimate the
value of a point or cell based on the surrounding input data and the
given variogram parameters. The two main algorithms are Simple
Kriging (known mean value) and Ordinary Kriging (unknown mean
value). The influence of the variogram parameters is very important for
the results.

Property Modeling Kriging • 97

Module 4 - Gaussian
Simulation

This module covers the basic concepts of Gaussian Simulation

techniques and how they are applied in Petrel. In this module, we will
focus on how to create 2D grids and the impact of changing the input
data and the variogram model settings.

Prerequisites
To successfully complete this module, the user must have knowledge of
the following:
• Fundamentals of Geostatistics, especially Kriging
• Petrel introduction

Learning Objectives
The purpose of this topic is for the user to understand the Gaussian
Simulation principle and the impact of the parameters in the result for
2D and 3D grids in Petrel.
At the completion of this module, you will have learned:
• The principle of Gaussian Simulation
• Comparison of Simulation vs. Kriging
• The influence of model parameter on Gaussian Simulation
algorithms
• Concept of unconditioned Sequential Gaussian Simulation
(SGS)
• Concept of Gaussian Random Function Simulation (GRFS)

Property Modeling Gaussian Simulation • 99

 Lesson

Gaussian Simulation
Simulation vs. Kriging
Kriging is a deterministic interpolation method giving the best local (and
smooth) estimate.
Gaussian simulation is a stochastic method based on kriging but capable of
capturing extreme values in a heterogeneous reservoir
Porosity

Simulation Kriging
Output • Multiple Equiprobable • One Deterministic Model
Realizations

Properties • Honors Wells

• Honors Histogram
• Honors Wells
• Honors Variogram ● Data samples (e.g. well data)
• Honors Variogram
Image • Noisy • Smooth away from wells
• Same variability Reality Simulation Kriging
everywhere
Use • Flow Simulation • Mapping
• Uncertainty Calculation • Volumetrics

Gaussian Simulation vs. Kriging

Reservoir modeling considerations
In reservoir modeling, the previously mentioned Kriging method is used
to calculate the best estimate (by minimizing error variance) of rock
properties, normally by interpolating well data. The limitations of such
a method is that in a large model, low values tend to be overestimated
and high values underestimated. In the search for oil-in-place in a
reservoir model, it is also important to capture extreme values.
Stochastic simulation methods are better at capturing the
heterogeneities (extreme value variation) of the subsurface by
assessing its (expected) spatial variability. Such stochastic methods can
use well data, in addition to relevant seismic or production data. They
can also be conditioned to existing facies models. Multiple realizations
can be run and uncertainty assessed.
100 • Gaussian Simulation Property Modeling
There are many ways to build a model with a result that is consistent
with the available sampled data:
• Estimation is accurate and smooth locally and is appropriate
for visualizing trends. It is inappropriate for flow simulation
where extreme values are important and does not assess
global uncertainty.
• Simulation reproduces histograms and honors spatial
variability (variogram). It is appropriate for flow simulation and
allows an assessment of uncertainty with alternative
realizations possible.
Simulation differs from Estimation (Kriging) in two ways:
1. Kriging provides a Best Local Estimate without considering the
resulting spatial variability. In simulation, global features and
statistics take precedence over local accuracy.
2. Kriging provides a single numerical model (BEST). Simulation
provides many alternative models (each one being a good
representation of the reality in some global sense).
Whatever procedure we choose to build our model with, the following
must be considered:
• Distribution of values
• Variability of value
• Connectedness of extreme values
• Our ability to assess the impact of uncertainty

Property Modeling Gaussian Simulation • 101

 Gaussian Simulation
Concept

Gaussian Simulation is based on Kriging and uses previously kriged /

simulated values as data to reproduce covariance between all simulated values
Gaussian distribution: Easy to establish conditional distributions (shape of all
conditional distributions = normal/gaussian)
Mean and variance: Given by Kriging

Sequential Gaussian Simulation is a widely used GSLib algorithm, also

used by Petrel for Stochastic Simulation. From a limited number of samples it
can build a model, considering the following:
Normal Score Transformation: Input data is transformed to give Mean=0 and
Standard deviation=1. The simulation results are back-transformed automatically
Stationarity: Mean of the data is assumed not to change laterally, spatial statistics do
not depend on locations (No trends in the data)
Output: Multiple EQUIPROBABLE realizations (organized by a random path)

Why Gaussian simulation?

The previously described Kriging technique is an estimation method and
is limited in the sense that the Kriging system forces the covariance
between the kriged estimate and the data value to be correct. But the
variance will normally be too small. In Gaussian simulation, the
covariance is kept unchanged, but a random and independent
component is added to correct the variance. Previously kriged or
simulated values are used as data to reproduce covariance between all
simulated values.
Gaussian or normal distributions makes it easy to establish conditional
distributions. This makes the SGS algorithm (see below) widely used for
its simplicity and mathematical stability.
Sequential Gaussian simulation (SGS) is the GSLib algorithm used
in Petrel for Gaussian Simulation. It requires Normal score
transformation of the data (Kriging does not require Normal score
transformation). SGS is a stochastic simulation that is used for sparse,
continuous data. It can honor input data, input distributions, variograms
and trends.
In the Make/edit surface process in Petrel, there are several options
for gridding up a surface. Using the SGS algorithm, Petrel automatically
102 • Gaussian Simulation Property Modeling
transforms data into Normal score. This is required by the algorithm and
the simulation results are back-transformed automatically.
In Property Modeling (SGS of Petrophysical Modeling), a Normal
score transformation is used automatically as well, unless you use the
transformations that have been done in the Data analysis process.
Many realizations can be run based on the same input data. However,
each realization will provide a different but equally probable result. The
realizations are defined by a semi-random path given by a starting point
in the matrix. This starting point is given by a seed number. If the seed
number is the same, the result will be reproduced.

Gaussian Simulation
Seed number – Decides the random path
Gaussian Simulation is stochastic by nature; i.e. many equiprobable
realizations can be created using the same input parameters, but varying
’randomly’ the visitation order of the unknown cells to be simulated
69069 2
2 11
11 8
8 9
9 14
14 7
Seed number 7 4
4 5
5 10
10 3

Cell number 3
16
16
1
1 6
6 15
15 12
12 13

The 3D grid is organized in Global Seed – used for all zones in a

numbered cells. The Random Path 3D grid
takes care of the visiting order Local Seed – used for individual Zones
/ Facies in a 3D grid

Gaussian simulation algorithms like SGS and GRFS will output

different realizations decided by a Seed number. The Seed number is
picked by a random number generator and randomly decides which cells
to visit first. The path is “random”, but “semi-random” in the sense that
if the seed number is used over again, the simulation will follow the
same path and create the same output (that is, an output can then be
recreated). Contrary to this, Kriging is calculating the cell values in
Property Modeling Gaussian Simulation • 103
 order and is only using the well data (not previously simulated values)
as input for estimating the unknown cells
The Petrophysical modeling process in Petrel has two options to
set up the seed number:
1. At the top of the Petrophysical modeling process dialog,
select the Seed check box and it will be used by the complete
model.
2. In the Distribution tab, select the Seed check box to set a
local seed to the zone of interest
The cell visitation order is decided by the seed number. However, in the
Expert settings of Property modeling, this can be changed to
visiting cells based on a property, in which case low values are visited
first.

Gaussian Simulation
Use of CDF curve and Variogram
Gaussian Simulation is dependant on the input (’hard data’) and the seed
number. It also requires a proper variogram setup (for range and anisotropy),
and a distribution to pick data from. The cell visitation order and output data
range (distribution) controls the shape of the CDF curve for back-transformation
of simulated values

The ’Core’
Within a given distribution the The ’Frame’
variogram parameters control
the spatial distribution

Before running a Gaussian simulation, the Variogram and

Distribution settings need to be specified.
The variogram provides the core, gives the range, azimuth, etc. which
the data will be simulated on.

104 • Gaussian Simulation Property Modeling

A CDF will be set up from the normal scored data. However, it is
possible to change the CDF by changing the output data range.
After simulation, when the data is back-transformed, it will be affected
by the change in the CDF curve, and a different result is provided from
the back-transformed data.

Gaussian Simulation
Process – step 1

1. The original distribution is transformed 2. At the grid point the Kriging Estimate and
to a Gaussian distribution using Normal Variance is calculated using neighboring
Score Transformation, giving a smooth data AND already simulated values
CDF curve

Variogram influence area

Principle of Gaussian Simulation - Step 1

1. The Histogram column (lower left plot) plots the porosity
values against frequency (how often they appear within one
bin/class). When the histogram is plotted as a cumulative
frequency, (top left plot), it gives a CDF curve. For Gaussian
simulation algorithms, a Normal Score Transformation (factor
in terms of Std. Dev.) of the data must be performed. This will
change the CDF curve to a smoother curve that the Gaussian
simulation algorithms will use.
2. Kriging (Simple/Ordinary) is used to calculate the unknown
value at a grid point using the neighboring data and the
already simulated data. Simple Kriging is used for calculating
the Standard deviation at the grid point. Generally, it is better
Property Modeling Gaussian Simulation • 105
 to use the Simple Kriging option because the second part of
the Kriging equation disappears since the mean=0. This is due
to the Normal Score Transformation (mean=0, std. dev.=1).
Simple Kriging is also faster than Ordinary Kriging.

Gaussian Simulation
Process – step 2

3. A Conditional cumulative distribution 4. The next grid node is visited following the
function (CCDF) is calculated based on random path. When the simulation is
original and previously simulated data. A finished, the value is back-transformed
simulated value is drawn from the CCDF. using the CDF curve of the input data.

Random
number

0
Simulated value

Principle of Gaussian Simulation - Step 2

3. More simulated values are provided during the simulation. The
CDF will be updated continuously, conditioned to the new
simulated values. This gives a continuously updated CCDF. A
simulated value is drawn (randomly using Monte Carlo
simulation) from the CCDF.
4. The random path decides which cell to visit next. After the
simulation, the data will be back-transformed from the
normal distribution to the original distribution
In the slide above, the red arrow indicates that a value will be picked
and simulated from the normal distribution.
The green arrow indicates that after the geostatistical algorithm has
been run, the picked value is back-transformed to its original
distribution.
106 • Gaussian Simulation Property Modeling
Theoretically:
Kriging (best local result) + Error fluctuation = SIMULATION
(best result in terms of global variability).
The fluctuations are simulated, and assuming an infinite number of
realizations, the simulation average will approach Kriging.

Gaussian Simulation
Gaussian Simulation Result – Compared to Input histogram

Gaussian Simulation result the shape of the histogram distribution of the

input data and the simulated result should be similar. The variability is preserved.

Input Histogram Simulation Histogram Gaussian simulation result

If the variability is preserved, the shape of the histogram distribution of

the input data and the simulated result will be similar.

Property Modeling Gaussian Simulation • 107

 Gaussian Simulation
Gaussian Simulation Result – Compared to Kriging
Gaussian simulation Kriging Input Histogram

Change-of-Support Effect

Kriging will smooth extreme values, and unlike Gaussian Simulation,

will not honor the input distribution. Therefore, there will be a
difference in the histogram caused by volume known as Change-of-
support effect.
The Change of support effect occurs when constructing geostatistical
reservoir models from, for example, well logs. One cannot ignore the
vast difference in scale when constructing the geostatistical model.
Spatial variance is important in the volume variance. It depends on
linear averaging (arithmetic averaging), and is correct for porosity but
not permeability.

108 • Gaussian Simulation Property Modeling

 Gaussian Simulation
Kriging/Averaged Simulation – Change of Support

Change-of-Support Effect
(Mean = 0.18; Stdev: 0.01/0.03/0.05)
Caused by different
volume support and the
120 smoothing character of
100 the (Kriging) algorithm
3D Grid
80
Frequency

60 Upscaled Cells
 Affine Correction
40 Raw Logs
s 2 (V 2)
20 z (V 1)corr = * ( z (V 1) − x ) + x
0
s 2 (V 1)
V1 Variable 1 (3D model)
0,05
0,07
0,09
0,11
0,13
0,15
0,17
0,19
0,21
0,23
0,25

Porosity (%)
V2 Variable 2 (upscaled cells)

Change of Support Effect

The supporting volume of a distribution affects the variance, but not the
mean. Coming from well logs via upscaled cells to the complete 3D
model, there is a tendency for less variability. This law is known and
the effect can be corrected.
Affine correction is a simple support effect correction. The variance
of a distribution can be reduced without changing the mean.
This can easily be done in the Property calculator:
Property_new=(Variance(Upscaled cells)/ (Variance(Property_
old))*(Property_old - Mean) + Mean
Example: A sand cell has an estimated porosity of 0.25. Furthermore, it
is known that the mean value for the sand is 0.20. The variance for all
upscaled sand cells is 0.0025 whereas the sand cells in the model have
0.0016.
Porosity_new=(0.0025/0.0016)*(0.25-0.20)+0.20=0.2781
After the correction the porosity in the sand cell is ~0.28.

Property Modeling Gaussian Simulation • 109

 Gaussian Simulation
Influence of the Variogram model parameters
Variogram Model Type (Range: 5000m)

Exponential and Spherical models Gaussian model gives

give quite similar results smooth result

The Gaussian variogram model honors all of the small values and
gives a smooth result.

Gaussian Simulation
Influence of the Variogram model parameters
Variogram Range (spherical variogram)

Range: 500m Range: 5000m

Nugget: 0 Nugget: 0

Simulation performed on large ranges gives more continuity in the

result.
110 • Gaussian Simulation Property Modeling
 Gaussian Simulation
Influence of the Variogram model parameters
Nugget (spherical variogram)

Nugget: 0 Nugget: 0.9

A small nugget honors the variability of the values, while a high nugget
gets a noisy result because every single cell will get a value with the
random path (for Kriging, the result gets a value close to the local or
global mean).

Property Modeling Gaussian Simulation • 111

 Gaussian Simulation
Influence of the Variogram model parameters

Anisotropy (spherical variogram)

Range: 20000m / 5000m Range: 20000m / 5000m

Azimuth: -45 Azimuth: 45

If the variogram search direction is oriented according to geological

know-how or a conceptual model, the result will be more geologically
meaningful. If not, the results of the existing data and a wrong
variogram may contradict the geological understanding of the area.
In the slide above, the azimuth on the left is the correct orientation of
the continuity.

112 • Gaussian Simulation Property Modeling

 Gaussian Simulation
Unconditional Simulation
In case NO INPUT data is available, an Unconditional Simulation can be performed:
User defines the output value range
User defines the mean and the standard deviation
The resulting grid gets the wanted value range via the back transformation from the
mean and standard deviation
The CDF curve is calculated to be used for back-transformation

1
CDF Curvature
defined by [Link].

Mean value Value range

0
Min. Max.

Unconditional simulation
Unconditional simulation can be performed when there are little or no
data available (like no upscaled logs).
In this case, running the simulation is mandatory to define the following
parameters:
Variogram parameters
Output range
Mean and Standard Deviation (Std. Dev.)
From the mean and Std. Dev. the CDF is calculated and used for
back-transformation.
Still, a CDF must be defined in addition to a variance. This can be
investigated by performing Data analysis.

Property Modeling Gaussian Simulation • 113

 Gaussian Simulation
Unconditional Simulation – Petrel
Make /edit surface process Petrophysical modeling process

No input or data set of

very few data points

Define the proper

Output data range

Define the CDF via

Mean and Std. Dev.

Gaussian Simulation
Unconditional Simulation Example

Sequential Gaussian Simulation:

Output range: 0 - 0.32
Mean: 0.15 and Standard deviation: 0.08

114 • Gaussian Simulation Property Modeling

Gaussian Simulation
Gaussian simulation algorithms available in Petrel

Sequential Gaussian Simulation

A popular stochastic method (GSLIB) based on Kriging and
error fluctuations. It can honor input data, input distributions,
variograms and trends.

Gaussian Random Function Simulation

It is faster than Sequential Gaussian Simulation, has better
reproduction of input statistics through parallelization, fast
Collocated co-kriging and additional expert settings.

Two Gaussian Simulation algorithms are available in Petrel:

Sequential Gaussian Simulation (SGS): It is a stochastic method
based on Kriging. It can honor input data, input distributions,
variograms, and trends. During the simulation, local highs and lows will
be generated between input data locations which honor the variogram.
The positions of these highs and lows will be determined by a random
number supplied by the user or the software. SGS has also been
improved in terms of speed through parallelization in 2010.
Gaussian Random Function Simulation (GRFS): It is faster than
Sequential Gaussian Simulation, has better reproduction of input
statistics through parallelization, fast Collocated co-kriging and
additional expert settings. The interface for the new algorithm is similar
to that of Sequential Gaussian Simulation, except for the Co-kriging
and Expert tabs.
For both algorithms, the Simple and Ordinary Kriging types are available
in the Petrophysical modeling process.
For more information about these methods, see the Petrophysical
modeling module: “Sequential Gaussian Simulation” and “Gaussian
Random Function Simulation algorithms”.

Property Modeling Gaussian Simulation • 115

 Gaussian Simulation – Exercises
Sequential Gaussian simulation is a stochastic method based on
Kriging. It can honor input data, input distributions, variograms and
trends.
During the simulation, local highs and lows will be generated between
input data locations which honor the variogram and the input
distribution. A random number, supplied by the user or the software,
will determine the positions of these highs and lows. Because of this,
multiple representations may be generated to gain an understanding of
uncertainty.
This exercise has a pre-made Petrel project stored in Projects folder:
Property Modeling [Link]. If you did the last exercise, please use
the same project.

Influence of variogram model parameters on Gaussian

Simulation
Exercise Steps
1. Display the point data set Ac Impedance in the 2D window.
2. Open the Make/edit surface process.
3. Enter a new Name for the output: “AI Simulated”. If there is
Before you display a name in the Result surface field, delete it. If a pop-up
the surface, open the window asks you to reset all settings, then click Yes to this.
Settings window for it (RMB
click on the “AI Simulated”
4. Select the Geometry tab and select the User defined option,
surface) and de-select the highlight the Ac Impedance data and click the Get limits
Show: Contour Lines option from selected button.
under the Style tab. This is
due to the random character 5. Set the Grid increments X and Y to 200.
of the simulation giving a 6. Go to the Algorithm tab and select the Sequential Gaussian
high number of contour lines. simulation as Method.
7. Leave the other settings as default and click Apply.

116 • Gaussian Simulation Property Modeling

 8. Display the surface on a 2D or Map window to visualize the
result.
9. Now, switch to Gaussian random function simulation as
method in the Algorithm tab.
10. Calculate different maps using the following variogram
parameters (defined under the Variogram sub-tab):
Range Major Range Minor Azimuth Nugget Model Type
Direction Direction
5000 5000 0 0.1 Spherical
500 500 0 0.1 Spherical
30000 30000 0 0.1 Spherical
30000 30000 0 0.9 Spherical
11. Observe the results on the AI Simulated surface that you have
displayed in the 2D/Map window.
When recalculating
12. Continue in the Make/edit surfaces process to calculate a the surface to get a different
surface using the variogram model that you determined output, the Result surface
previously in the Sample variogram exercises (Module 2 Basic must be removed by
selecting it, Delete and to
Statistics Part 2 - Exercises). give a different Name (“AI
13. Leave the previous settings, but remove the Result surface Simulated GRFS”). A pop-up
window ask to reset all the
and enter a new Name. settings, select No to keep
14. Select the Variogram model from the Input pane (“Sample the old parameters and only
Var. from Ac Impedance (-39 deg)”). apply the changes needed.

Property Modeling Gaussian Simulation • 117

 15. Go to the Algorithm tab, select Sequential Gaussian
simulation, in the Variogram sub-tab, click the
button to get the parameters from the variogram model.
16. Click Apply/OK to calculate the new surface and display it to
compare with the previous results.
17. Keep all parameters, but change the Seed number under
Settings sub-tab>Specify seed, and recalculate the surface.

Optional Exercise: User defined Normal distribution for

simulation
Prior to Gaussian Simulation, the data is transformed to Normal
distribution and the simulation results will be back-transformed. You do
have limited control over the back transformation in the Settings
folder:
• Output data range - is typically the data range of the input
data. The CDF used for the back transformation covers the
value range defined by this parameter.
• If you select the check box Distribution from input data, then
the CDF used for the back transformation is calculated from the
input data.
• Normal distribution – use this function if you have few or no
data points and want a surface with normally distributed data.
Using the Output data range (Min. and Max.), Mean, and
Std., you control the data range of the output. Use this option
with great care, as it will have a massive influence on your
result and you may lose all relationship with your input data.
Make sure the Mean is in the middle of the data range. Also,
the data range should be larger than Mean+/- Std. Deviation.

Exercise Steps
1. In the Make/edit surface process, delete the Result surface
and select Yes in the pop-up dialog asking if you want to reset
the settings.
2. Enter an output Name: “User def distribution”.
3. In the Geometry tab, activate the “AI Simulated” surface
from the Input pane and click the Get limits from selected
button.
118 • Gaussian Simulation Property Modeling
 4. Go to the Algorithm tab and choose Sequential Gaussian
simulation.
5. In the Variogram sub-tab, choose an isotropic variogram
model with a range of 5000 m and a nugget of 0.1.
6. In the Settings sub-tab, define the following parameters:
• Choose an Output data range of –1000 / +1000 and
select Absolute.
• Select Normal distribution.
• Select a Mean of 0 and a Std of 10.

7. Click Apply to calculate the surface. In the surface

Settings>Style tab, turn off the Contour lines Show check
box before displaying the surface. Also Adjust color table on
selected button from the main toolbar.
8. Change the Std to 1 and recalculate the surface. Refresh the
color table and observe the result.

Property Modeling Gaussian Simulation • 119

 Summary
In this module, you have learned that stochastic simulation techniques
(Gaussian Simulation methods) incorporate Kriging algorithms and a
random residual component to simulate and generate multiple equally
probable results. This is different from Kriging as the already simulated
values are used in the estimation of new cells values.
You also learned that a random path (provided through different seed
numbers), guarantees different equiprobable outputs, and that the
influence of the variogram parameters on the final result is important.
Also, the possibility of Unconditional Simulation (no input data) is a
great advantage of these algorithms.

120 • Gaussian Simulation Property Modeling

Module 5 - 3D Property
Modeling Overview

This module provides a brief overview of the different tools available in

Petrel for 3D Property modeling. The basic concepts and algorithms of
Property modeling will be explained together with the software
functions.

Prerequisites
No prerequisites are required for this module.

Learning Objectives
The purpose of this topic is to give the user a general understanding of
3D Property modeling. At the completion of this module, you will be
able to:
• Understand the Petrel workflow tools for 3D Property modeling
• Realize the differences between deterministic and stochastic
modeling techniques

Property Modeling 3D Property Modeling Overview • 121

 Lesson

Petrel workflow tools

The above slide shows the Petrel workflow tools which are available
for Property modeling which give you the ability to:
1. Import all of the available data for your project.
2. Analyze and quality check the initial data for your property
model.
3. Prepare a proper data set (look at other options like Neural
Networks)
4. Use the 3D structural model (geocellular model) for distribution
of properties
5. Data analysis for discrete and continuous properties (Upscale
logs and quality check them)
122 • 3D Property Modeling Overview Property Modeling
 6. Build a realistic 3D Property Model, populating the model with
facies and petrophysical properties using deterministic and
stochastic methods (according to the data amount)
7. Make volume calculation and uncertainty analysis (for
evaluating different scenarios)
8. Upscale the static model properties for simulation
9. Simulation to predict production
This module provides only an overview of 3D modeling techniques and
workflows in Petrel.
In the following modules, you will learn how to prepare a proper
dataset for Property modeling. Well logs will be upscaled into the 3D
grid cells that are penetrated by a well using the Scale up well log
process.
Using the Facies modeling process, the statistical discrete data
analysis and how to make 3D models with discrete data will be
introduced.
Petrophysical data analysis, including how to transform the data and
how to create variograms from the data, will be shown before using the
Petrophysical modeling process to create different petrophysical models
of porosity, water saturation, net to gross and permeability.
How to quality check both the input data and the resulting model will
be stressed during this course.
The basic concepts and algorithms of Property modeling will be
explained together with the software functionality. The course focuses
on the practical use of Petrel functionality for performing property
analysis and distribution. 

Property Modeling 3D Property Modeling Overview • 123

 3D Property Modeling
Facies Modeling methods – Petrel
Deterministic Learning system
Estimation Direct Addressing Artificial

Indicator Kriging Assign values Interactive Neural Net

Discrete distribution of Values can be Allows the user to paint Uses the classification
the property honoring undefined, constant, facies directly on the model made in the Train
the pre-defined another property, 3D model Estimation Model process
histogram. surface or vertical
function.

Facies modeling
Deterministic algorithms
Deterministic methods will always give the same result with the same
input data. These algorithms will generally run much quicker and are
very transparent - it is easy to see why a particular cell has been given
a particular value.
The disadvantage is that models with little input data will automatically
be smooth even though evidence and experience may suggest that this
is not likely. It is often difficult to perceive the uncertainty of the model
away from the input data points. Consequently, interpolation based
deterministic methods are generally too smooth, and can lead to bias.
Usability
Deterministic techniques are typically used when dense data is
available (for example, many wells, wells + seismic). The deterministic
methods yield a single estimated result (such as they do not produce
multiple realizations) and cannot account for the uncertainty of a model.

124 • 3D Property Modeling Overview Property Modeling

For Estimation:
Indicator kriging - allows a discrete distribution of the property,
honoring the pre-defined histogram.
For Direct addressing:
Assign values - Gives you five different options to choose from:
Undefined, Constant, Other Property, Surface and Vertical Function. For
each of these, you can keep or overwrite the original values of the
upscaled logs.
Interactive - Allows you to paint facies directly on the 3D model. The
Facies painting tools are stored in the function bar for Facies modeling.
For Learning systems:
Neural net - Uses a classification model made in the Train estimation
model for creating a discrete property.

3D Property Modeling
Facies Modeling methods – Petrel (cont.)
Stochastic
Pixel based Object based
Sequential Truncated Truncated Multi-point Facies Object Modeling
Indicator Gaussian Gaussian Simulation • General object
Simulation Simulation Simulation with • Fluvial
trends • Adaptive

Honors input Used where facies Distributes facies A training image Allows population
distribution, and are known to be based on transition describes the of a discrete facies
directional settings sequential. Honors between facies, spatial correlation model with different
like variograms and input data, global and trend direction from one-to- bodies of various
trends. fractions and (converted into multiple points (for geometries, facies
trends. probabilities). facies and their and fractions.
relative position).

Property Modeling 3D Property Modeling Overview • 125

 Stochastic algorithms
Stochastic methods are good at incorporating secondary information
such as seismic data. The best stochastic models are the ones that
make maximum use of any and all relevant observed data.
Stochastic methods use a random seed in addition to the input data, so
while consecutive runs will give similar results with the same input
data, the details of the result will be different. Stochastic algorithms
are more complex and will take longer to run. However, they do honor
more aspects of the input data, specifically the variability of the input
data. This means that local highs and lows will appear in the results
which are not steered by the input data and whose location is purely an
artifact of the random seed used. The result will have a distribution
more typical of the real case, although the specific variation is unlikely
to match. This can be useful, particularly when taking the model further
to simulation as the variability of a property is likely to be just as
important as its average value. The disadvantage is that some
important aspects of the model can be random and it is important to
perform a proper uncertainty analysis with several realizations of the
same property model with different random seeds.
The stochastic models produce realizations which fluctuate close to the
correct mean (provided the model roughly represents the true reservoir
heterogeneity). The interpolated models will systematically bias the
results
Usability
Stochastic techniques are often used in conditions where sparse data is
present. These methods produce a possible result and can be used to
produce multiple equally probable realizations.
In this course, we will concentrate on stochastic techniques.  
The current suite of Facies modeling techniques that can be used in
Petrel are explained below. They range from interactive facies modeling
(paint brush approach) to various stochastic methods capable of
generating multiple realizations. The most widely used methods are
Sequential Indicator Simulation (SIS) and Object Modeling (stochastic).
Object modeling (stochastic) allows you to populate a discrete
facies model with different bodies of various geometry, facies code and
fraction.
126 • 3D Property Modeling Overview Property Modeling
Sequential Indicator Simulation (SIS) allows a stochastic
distribution of the property, using the pre-defined histogram. Directional
settings, such as variogram and extensional trends, are also honored.
Truncated Gaussian Simulation (TGS) is a fast modeling technique
for discrete properties, where the facies are known to be sequential
and the variograms defining each of the facies are the same. It is most
commonly used in carbonate environments, and the method is pixel
based; hence, it can easily deal with large amounts of input data,
specified global fractions, 1D, 2D or 3D trends.
Truncated Gaussian with trends (TG with trends) allows a
stochastic distribution of the facies based on a given transition
between facies and a trend direction. These trends are then converted
into probabilities and the model is run using the standard Truncated
Gaussian technique, which allows you to specify global fractions for
each of the facies. The trend shape and direction are set interactively in
the dialog and a range given for the variogram.
Multi-point facies simulation (MPFS) is a pixel-based algorithm
based on Multi Point Statistics (MPS) technique and creates facies
models which may look like object models. The principal idea of MPS is
the description of the spatial correlation from one-to-multiple points at
the same time. As opposed to the two-point statistics (point - to - point
like Kriging or Stochastic Simulation), the variogram will be replaced by
a training image (2D or 3D). This training image describes the
geological facies in relative position to each other.

Property Modeling 3D Property Modeling Overview • 127

 3D Property Modeling
Petrophysical Modeling methods – Petrel

Deterministic
Estimation Interpolation
Kriging Kriging Kriging by Closest Functional Moving
Interpolation GSLIB average

Honors well Handles large Standard It uses the Honors well Based on the
data, input datasets and GSLIB method closest well and trend data input it gives an
distributions, works both in with option of data input for using a 3D average value
variograms and real collocated co- each function for the and calculates
trends. It can coordinates kriging. Works unsampled interpolation. the weigths
work in real (X,Y,Z) and grid only in grid location. according to
coordinates coordinates coordinates the distance
(X,Y,Z) and is (I,J,K ). Fast (I,J,K). from wells.
fast. Collocated co-
kriging.

Petrophysical modeling
Deterministic techniques
Typically used when dense data is available (e.g., many wells, wells +
seismic) or when modeling properties of more continuous response. The
deterministic methods yield a single estimated result.
For Estimation/Interpolation:
Kriging by Gslib (estimation) - The algorithm is supplied by GSLIB
and has an option for collocated co-kriging. It works in IJK space only
and can handle simple and ordinary kriging.
Kriging interpolation (estimation) - Honors well data, input
distributions, variograms and trends. Values between input data points
are estimated deterministically using the variogram.
Kriging (estimation) - Honors well data, input distributions,
variograms and trends. It has some additional expert settings, and you
can switch between simple and ordinary kriging. It differs from the
standard GSLIB Kriging as it is able to krige a much larger number of
128 • 3D Property Modeling Overview Property Modeling
points through a combination of multithreading and smart neighborhood
searching. It has considerable performance improvements through
parallelization, the addition of a fast Collocated co-kriging algorithm
(implemented as an interactive slide correlation coefficient slide bar),
the ability to model in XYZ and IJK space and some additional settings
in the Expert tab.
Closest (estimation) - Honors well data. Uses the closest input point
for each unsampled location.
Moving average (interpolation) - Honors well data and trends. Takes
an average value for each unsampled location based on input data, and
calculates weights according to distance from wells.
Functional (estimation) - Honors well data and trends. Creates a 3
dimensional function (Parabolic, Simple parabolic, Planar, Bi-linear) to
be used in the interpolation.
Deterministic-direct addressing
Assign values - A group of options including the possibility to give all
the cells in the selected zone a constant value, an undefined value,
keep previous values or sample values from another property, vertical
function or surface. A seismic volume can also be sampled into a
property (using Geometrical modeling). In addition, the property
calculator can be used to assign fixed values. (see next slide)
Learning system
Neural net - An estimation model made in the Train Estimation process
can be used for a continuous property generation. Use a combination of
the Neural Net method (Petrophysical modeling) and the Train
Estimation Model process. (see next slide)

Property Modeling 3D Property Modeling Overview • 129

 3D Property Modeling
Petrophysical Modeling methods – Petrel (cont.)

Deterministic Learning system Stochastic

Direct Addressing Artificial Pixel based
Assign values Neural Net Sequential Gaussian Gaussian Random
Simulation (GSLIB) Function Simulation

Choose from undefined, Uses the Estimation Honors well data, input It is faster than SGS,
constant, other property, model made in the Train distributions, variograms and gives better
surface and vertical Estimation Model and trends. The variogram reproduction.
function. process. variogram and It has a fast collocated
distribution are used to co-simulation option.
create local variations
away from input data.

Stochastic techniques
Typically used in conditions where sparse data is present. These
methods can produce multiple equally probable realizations that helps
us understand the degree of uncertainty. Petrophysical data can still be
very constrained by conditioning them to existing facies models.
This course concentrates on the Sequential Gaussian Simulation (SGS)
and the Gaussian random Function Simulation (GRFS) methods.
Sequential Gaussian simulation (SGS) - Honors well data, input
distributions, variograms and trends. The variogram and distribution are
used to create local variations, even away from input data. As a
stochastic simulation, the result is dependant upon the input of a
random seed number and multiple representations are recommended to
gain an understanding of the uncertainty.
Gaussian random function simulation (GRFS) - Honors well data,
input distributions, variograms and trends. This algorithm provides
much the same output, but differs substantially in other aspects from
130 • 3D Property Modeling Overview Property Modeling
the SGS from GSLIB. It is typically faster than SGS, it is a non-
sequential algorithm (parallelized) and has a fast collocated co-
simulation option. The unconditional simulations use a Fast Fourier
Transform based algorithm that is fast and gives a good variogram
reproduction for a wide class of variograms.

Property Modeling 3D Property Modeling Overview • 131

 Summary
In this module, a short introduction to different deterministic and
stochastic modeling techniques was given.
The Petrel workflow tools for 3D Property modeling were also
introduced.

132 • 3D Property Modeling Overview Property Modeling

Module 6 - Property
Modeling Data Preparation

In this module, you will learn how to prepare the input data for Property
modeling and how to generate discrete and continuous logs
(lithological/ sedimentological facies and petrophysical properties)
interactively in a Well section, by using the Well calculator and the
Train estimation model (Neural Networks).
In Facies modeling, we will focus on facies logs preparation. For the
Petrophysical modeling, a dataset has been prepared. The project
contains porosity and permeability logs. In addition, we will show you
options for preparing water saturation and net to gross logs.
Although not covered in this course, the facies and petrophysical
properties can be imported from other applications.

Prerequisites
To successfully complete this module, the user must have knowledge of
the following:
• Familiarity with Geology Fundamentals: Basic Sedimentology
• Familiarity with Petrophysics Fundamentals
• Well Correlation
• Petrel Introduction

Learning Objectives
The purpose of this module is to learn the different processes used to
generate Facies and Petrophysical logs in Petrel. At the completion of
this module, you will be able to:
• Make interactive facies interpretation in a Well section
• Use the Well log calculator to create facies logs with the
defined cutoffs
• Use the Train estimation model process to create a facies log
• Use the Well log calculator to create water saturation and net
to gross logs with the defined cutoffs
Property Modeling Property Modeling Data Preparation • 133
 Lesson

Property Modeling Data Preparation

Initial dataset
Well / Seismic Data
Facies logs
– lithology
– rock types
– sedimentological facies etc.
Petrophysical logs / core data
– mineral volumes
– porosity & permeability,
– water saturation, net pay etc.
Secondary data,
– seismic attributes (related to facies or petrophysics)

Reservoir modeling
Modeling requires a good geological understanding of the reservoir, an
apprehension of the quality of the available data and a good
understanding of the different modeling techniques.
The data preparation process includes importing well data and
seismic attribute cubes and preparing the data for use in the Property
modeling. You will be using certain processes such as the well
calculator, well log editor and well correlation.
The preparation also includes defining a conceptual sedimentological
model and an appropriate geological knowledge of the area.

134 • Property Modeling Data Preparation Property Modeling

 Property Modeling Data Preparation
Facies / Lithology interpretation
Facies or lithology interpretation can be obtained from:
Crossplots
Individual or combined log readings (log quicklooks)
Known depositional environment or field analogs
Cuttings report
Techlog interactive software suite

Property Modeling Data Preparation

Facies / Lithology interpretation in Petrel
Facies/Lithology or discrete properties can be imported or created by:
• Log Calculator – creating cutoffs based on existing logs
• Interactive drawing – using paintbrush option from Well section
• Artificial Neural Networks – classify discrete logs based on various input logs
• Import – using ASCII files, via OpenSpirit or Techlog Link plugin
• Synthetic logs – generated from 3D model properties

Property Modeling Property Modeling Data Preparation • 135

 Facies / Lithology Interpretation
Discrete logs can be interpreted by using the Well log calculator,
through interactive sedimentological facies interpretation or a
combination of both. The normal workflow is to do a rough generation
of a facies log by using the calculator. Afterwards you do a more
detailed interpretation based on all of the well logs by using the
interactive sedimentological facies interpretation.
The discrete logs can be imported from other databases or they can be
created by using the Train estimation model process (Neural net
Classification method).
Interactive interpretation in a Well section: Display the well logs
that give information about the facies and then draw the different
facies in the correct position in a separate track.
Calculator: As an example, the calculator can be used to set an
expression such as < If the Gamma ray logs value are less than a
certain value then set facies to sandstone, otherwise set facies to
shale>.
Train estimation model: This process gives you access to tools for
Neural Network analysis enabling you to create a classification model
object (for example facies).
Import: Petrel can read LAS or ASCII for manually importing or via Open
Spirit automatically.
OpenSpirit: Petrel 2010 supports the OpenSpirit application which
works as a middle layer between software applications and various
databases and data repositories, allowing Petrel to import and export
data to the database, irrespective of the platform.
Techlog Link Plug-in: If you are using Petrel and Techlog*, a Techlog
Link plugin is available for Petrel that can transfer directly data like
logs, borehole images and tops between Petrel and Techlog. This plugin
will dramatically ease the dataflow between the two Windows
applications without the need to use third party applications, files or
other ad-hoc methods.
[*] Mark of Schlumberger.

136 • Property Modeling Data Preparation Property Modeling

 Property Modeling Data Preparation
How to create a Facies template
If continuous logs are imported (like GR, SP etc.), these logs can be used to create
basic Facies / Lithological logs. First we must define a new Facies template:

1. In the Templates pane, expand and

1
right-click the Discrete property
templates folder.

2. Select Insert new property template. 2

3
3. Open the new ’Untitled 1’ template and
5
use the remove/add rows buttons.

4. Enter facies names and change color 4

and pattern under the Colors tab.

5. Go to the Info tab and rename the

template.

Property Modeling Data Preparation

Create new Facies log using the Well logs calculator

Create a new discrete log that is defined by cut-off values from a continous log:

1. Use a log Calculator: 2

from the Wells folder or 5
individual wells.
3
2. Enter a new log name
and the expression. 4

3. Select the Property

template you just made.

4. Click ENTER.
5. The new log is stored
in the Global well logs
folder. Compare in Well
section window.

Property Modeling Property Modeling Data Preparation • 137

 Property Modeling Data Preparation
Create new Facies log by doing interactive Facies interpretation

Create a new discrete log using

an interactive paint option:

Create discrete logs

Paint discrete log class

+
Create new discrete log

Edit discrete logs

Paint discrete log class

Flood fill discrete log class

Uncertain area, only 1
main sand interpreted
Pick up discrete log class

Property Modeling Data Preparation

Neural Networks – Method and Input Data

Aim: Create a new classification discrete log based on various continuous input logs in key
wells. Each input log can be correlated to establish a general relationship.

Choose Model type: Classification

for discrete logs and Estimation for
continuous logs.

Select Data Type: well logs.

Select Wells to be used; these

should have some logs in common.

Select Logs to be used as input

data for Training.

Press Correlation analysis to see

how selected logs inter-relate.

138 • Property Modeling Data Preparation Property Modeling

Artificial Neural Networks
An Artificial Neural Network is a computational model based on
biological neural networks. It consists of an interconnected group
(network) of artificial neurons (nodes) and processes information by
using a connectionist computation approach (interconnected networks
of simple units).
In most cases, a Neural Network is an adaptive system that changes its
structure based on external or internal information that goes through
the network during the learning phase.
A Neural Network is a non-linear statistical data modeling tool, able to
model complex relationships between inputs and outputs or to find
patterns in the data.
Example of use in Property modeling - Predict logs:
By identifying the relationship between facies and other measured logs
like Sonic, GR and resistivity in a few key wells, this relationship can be
used in any other well where the input logs are present to make an
estimate facies log.

How to setup a Neural net to create a facies log

Activate the Train estimation model process in the Processes pane
> Utilities.
Our aim is to look at some wells containing basic continuous log data,
use these data and their correlations to create a discrete output that
can be compared to the interactive drawing we did earlier. For example,
two discrete classes as output:
1. To do a simple Classification, select a few wells that contain
some basic logs (for example GR and Porosity).
2. The log data the chosen wells have in common will have an
active check box, while the remaining logs will be grayed out.
3. Note that the correlation that is output should not be too
close to 1 as that indicates the data may come from the same
source. However, if the correlation is too low, there is no
relationship at all between the data and it will be difficult to
achieve a good mathematical model that can be used for
prediction.

Property Modeling Property Modeling Data Preparation • 139

 Property Modeling Data Preparation
Neural networks – Classification (create Facies log)

If one does not know what to expect from the output but want to classify data into ’bins’, an
unsupervised classification run is useful. If a target log (discrete) of higher confidence zone data
has been made, a supervised classification run is more useful.

Unsupervised
As a first pass use Unsupervised training
and specify the amount of classes to output

Supervised
Supervise the log with an existing Facies Original
log. Select Supervised and a discrete log. Facies log

Resulting Well Logs Unsupervised

1. Training image log
2. Unsupervised log (general discrete
template)
3. Supervised log (inherits the training Supervised
image log template)

4. Go to the Settings tab, try to use Unsupervised as a first

approach (if no facies logs exist), specify the amount of classes
to output (here 2, possible sand and shale).
5. Use the default Training parameters. Click Apply/OK to run
the process.
The training result is the Classification model which can be used for
prediction. The purpose of the Neural Net method is to create a general
relationship between some input parameters. This general relationship
is called an Estimation Model. Once a relationship has been
established, and has been validated as a “good general relationship”, it
can be used for prediction at unknown points.
6. Select the new General Discrete (Neural net 1) log from
Global Well logs folder. Compare it to the input logs and the
interactive facies interpretation done earlier.
7. The objective facies code can be changed to specific names
and colors, resembling for example your own interpretation.
This can guide you further, and is applied to all wells.

140 • Property Modeling Data Preparation Property Modeling

 Property Modeling Data Preparation
Neural Networks – Training Parameters and Output
As the NN is based on a training rule, some training parameters must be set up. These decide how
many times the training is run, data selection /split and the stop criteria. The ouput is a log and a
Neural net output containing input logs (Attribute) and ouput classes (Class filter).

Training Parameters
In a first pass, leave the training parameters as default.
These can be tuned later if the output is undesirable.

Well Log
The result yields a log under Global well logs.

Training Result
The result creates a new output called Neural net 1 in
the Input pane – this shows the input attributes and the
output classes.

Note: The amount of classes is decided by the

number of classes defined or the property template
used by the supervised log.

Classification workflow summary

Step 1: Establish training model based on key wells, providing input
data for a Neural net model
Step 2: Use input logs that have some relation to your desired output
(for example clay volume, effective porosity and permeability).
Step 3: Instead of defining a set of unknown facies classes
(Unsupervised training), the key facies intervals can be identified from a
core study. The intervals are “drawn” as a facies interpretation (desired
output) with a set of known facies classes; then Supervised training will
be performed.
Step 4: The Neural net method is performed (Supervised training) and
results in a computed output (facies log)
Based on the Training parameters, the errors found between the desired
output and the computed output in iteration 1 is back-propagated as
weights to adjust the training system. The training parameters will
decide on how many iterations will be performed.
Property Modeling Property Modeling Data Preparation • 141
 Property Modeling Data Preparation
Petrophysical parameters
For a Petrophysical Evaluation the common parameters are:

Porosity - fraction of a unit volume filled with pores or voids

Total porosity φt = Total pore space / Bulk volume

Effective porosity φe = Interconected pore space / Bulk volume

Water Saturation - fraction of pore space occupied by

water. Initial SW is used for calculation STOIIP.
Water saturation Sw = Sw irreducible + Sw free

Permeability - dynamic property associated with fluid

movement . Permeability is sometimes related directly to
Porosity, but not always (a low porosity carbonate may have
a “high” effective permeability due to extensive fractures).

Property Modeling Data Preparation

Petrophysical parameters (cont.)
Net to Gross – fraction of the potential reservoir as a ratio of the total volume

Example:
Gross thickness= Total interval thickness=2345-2252= 93 m

Net thickness = Total thickness of sand= 72 m

Net to Gross = Net / Gross=72 / 93= 0.774

Pore thickness = Net thickness * Porosity= 72*0.32= 23.04 m 93 m

Hydrocarbon pore thickness = Pore thickness * (1-Sw)

= 23.04*(1-0.2)= 18.432 m

142 • Property Modeling Data Preparation Property Modeling

Additional notes on Net-to-Gross (NtG) and Modeling
NtG may not be used at all in your 3D model, but sometimes, especially
with lack of facies data or a detailed facies model, NtG can be used to
distinguish the “good” part (for example, sands) from the “bad” part (for
example, silt/shale). It is most commonly used for clastic depositional
environments and not so much for carbonates.
Net-to-Gross at well log level
Net Reservoir - is the potential reservoir interval containing
hydrocarbons. It is generally a function of Vclay/Vshale and Effective
Porosity
Net Pay - is generally seen as the producible part of the Net reservoir
and is a function of Vclay/Vshale, Effective Porosity and Irreducible
Water saturation. We can also call this Net-to-Gross, but people use
these terms with varying meaning, so it is a good idea to always check
which parameters they involve and that the correct input and use is
communicated.
When applying cutoffs for logs the Result is a flag value of 1 and 0 at
each log sample point (1 = Pay zone, 0 = No pay).
Net-to-Gross in 3D modeling
Upscale binary (1 and 0) continuous logs, or another continuous log so a
NtG distribution in Upscaled position is used as basis for 3D modeling.
Model NtG conditioned to Facies, also where there are sections of
no-reservoir, to avoid bias.
Net-to-Gross in Petrel Volume Estimation
• If you do not condition your property values to a facies, you
have to use a NtG value which is associated with the entire
interval for the zone.
• When you condition the property to a particular facies model,
use 1 as NtG in the volumetric calculations; by creating the
facies model and conditioning the property to the facies, the
Net-to-Gross is already accounted for.
• You may want to use cutoffs in the Property calculator to
calculate a NtG property which is dependent on the facies; for
example If facies = sandstone, then net/gross = 0.6, If facies =
silt, then net/gross = 0.3. However, this is a hard constraint
that will not allow variations around the model as conditioning
or Co-simulation will.
Property Modeling Property Modeling Data Preparation • 143
 Property Modeling Data Preparation
Petrophysical logs – Interpretation in Petrel
Petrophysical or continuous properties can be created or imported by:
• Log Calculator – creating cutoffs based on existing logs
• Artificial Neural Networks – estimate continuous logs based on various input logs
• Import – using ASCII files, via OpenSpirit or Techlog Link plugin
• Synthetic logs – generated from 3D model properties

Petrophysical Log Interpretation

Continuous logs can be interpreted by using the Well logs calculator.
These logs can also be imported from other databases or they can be
created by using the Train estimation model process (Neural net
Estimation method).
Calculator: The calculator can be used to set up a different expression
or pre-defined formulas to calculate petrophysical properties.
Train estimation model: This process gives you access to tools for
Neural Network analysis enabling you to create an estimation model
(e.g porosity or permeability logs).
Import: Petrel can read LAS or ASCII for manual imports or
automatically via Open Spirit or Techlog Link plug-in.

144 • Property Modeling Data Preparation Property Modeling

 Property Modeling Data Preparation
Well Log Calculator - Sw calculation example (Archie’s equation)
Archie’s equation is the basic equation used for logs by Petrophysicists to determine
whether a formation has water or hydrocarbons in the pore space (Water Saturation-Sw)

Well log calculator

Enter the formula to
calculate a new SW log.
The different variables
should be defined with a
constant or log.

a Rw
Sw = n
φ
m
Rt

Note: a, m, and n depends on lithology.

Archie’s Equation for Water Saturation (SW) logs

An accurate determination of the initial oil in place (STOIIP) in the early
life of reservoirs or an evaluation of the developed reservoir is required
to estimate the hydrocarbon volumes in place or left in the reservoir. A
modified Archie formula is the basic equation to compute water
saturation in clean formation. The exactness of the water saturation
value for given reservoir conditions depends on the accuracy of the
Archie parameters a, m and n.
There are several saturation equations, such as: Indonesia Equation,
Nigeria Equation, Waxman-Smiths Equation, Dual-Water Equation. All
reduce to Archie’s equation when there is no shale.
As shown in the slide the Archie’s equation water saturation is a
function of several variables:
Rw - resistivity of the water in the pore space
Rt - total composite formation resistivity
Ø - Porosity

Property Modeling Property Modeling Data Preparation • 145

 Other variables that are a function of the rock
n - saturation exponent (~2). It can be set up to 2 in the simple case,
higher in oil wet rocks and lower in water wet rocks
m - cementation exponent (~2). It can be set up to 2 in the simple case,
>2 in poorly connected moldic formation or <2 in fractured formation
a - empirical constant (~1). It can be set up to 0.8 in sandstones and 1
in limestones

Property Modeling Data Preparation

Well Log Calculator - Sw calculation example (J-Function)
The ”J- Function” relates Sw to capillary pressure (Pc), permeability (k) and porosity (Φ).

1. Make a J-log, based on the permeability and porosity logs available by using the
Global well log calculator.
2. Check the different Sw lookup curves in a Function window; they are normally
different for different lithologies/rock types.
3. Use the J-log to calculate SW log.

Pc k Sw irr.
J (S w ) =
s φ SW = a x J b

Note: a and b will affect the curve; depends on lithology.

J-Function for SW logs

The calculation of SWI (initial fluid saturations) is important as it
influences the initial oil in place (STOIIP) estimation in any 3D reservoir
model. Subsequently, it affects the dynamic modeling (history match
and predictions). If cores are available, combined core data permeability
with identifiable reservoir rock types (RRTs) and log data can be used to
find capillary pressure (Pc). Then a J-function can be calculated for each
RRT subsequently used to calculate the SWI in the reservoir.

146 • Property Modeling Data Preparation Property Modeling

Water saturation is based on net effective porosity which is the
average effective porosity of the “producible” part of a rock unit.
The J-function (Leverett) should be calculated differently for Oil and
Gas as the input parameters change.
Example:
J-oil = SQRT(Perm/Poro) x deltadensity (=0.301 between oil to water) x
3.14 x 9.81 x H_above_contact
J-oil = J-oil / (interfacial tension x Cos of contact angle x 100) = J-oil /(
30 x 1 x 100)
Sw = a J exp b = 0.347 x J exp –0.486
J(Sw) = J-Function for capillary pressure (Pc)
Capillary pressure occurs in porous media when two or more immiscible
fluids are present in the pore space; it is the pressure difference Water saturation can
between the wetting and the non-wetting phase. Since gravity forces be calculated from logs,
upscaled and populated in
are balanced by capillary forces, Pc at a point in the reservoir is the 3D grid. However, many
estimated from the height above the oil-water contact (or oil and water prefer to get the water
phases) and the difference in fluid densities. Pc curves are normally saturation in as a separate
function and use it above a
estimated in a laboratory. given contact directly in the
3D grid. On the next slide we
Reservoir calculations require a normalized curve for the Pc will show you how to model
measurements and are obtained from several plug samples at different SW directly in the 3D grid.
depths. Due to the heterogeneity of a reservoir, no single curve can be
derived.
Leverett introduced the first dimensionless Pc correlation function
J(Sw). This function accounts for changes in Perm, Porosity and
wettability when the pore geometry is constant.

Property Modeling Property Modeling Data Preparation • 147

 Property Modeling Data Preparation
Property Calculator - Sw property calculation example (from file)
In this example, Sw is a function of height above fluid contact, permeability, and porosity.

1. In the Property calculator enter an expression.

This requires: Above Contact property from
Geometrical modeling and Permeability and
Porosity from Petrophysical modeling.

2. Or, load a macro where the expression is written

to simplify the calculation (.mac/.txt file):
In the Property calculator, select the From file
check box, browse and Run.

Sw calculation using the Calculator for properties

1. Create an Oil-Water contact under the Make contacts
process.
2. Go to the Geometrical modeling process, select the method
Above contact and insert the Oil-Water contact. Rename the
new property to Above_Contact.
3. In the Input pane, insert a new folder and name it Functions.
Right-click>Create new function for the height of water
saturation inside Petrel or in the same folder Right-
click>Import on selection>select Function X,Y (ASCII) format
to import a pre-existing function file (the function should have
one vertical column with height and one with the Sw values).
4. Use the Height above contact to make a Sw property using the
function. In the Properties folder>right-click>select
Calculator and type in Sw=Function_name(Above_
Contact) and use Sw template to make a new Sw property.
148 • Property Modeling Data Preparation Property Modeling
Sw calculation using Permeability/Porosity relationship
(J-function) from a macro file
Create your statements in a text editor (macro) or in the Property
calculator. For example:
Temp=(AboveContact)*SQRT(Perm/Poro)
Sw=0.5*Pow(Temp,-0.2)
Sw=IF(Sw>1,1,Sw)
Sw=If( Sw=U,1 , Sw)
If the statement is created in a text editor, make sure it has a .mac/.txt
extension; then select From file in the Property calculator, browse the
file (...) and click RUN.

Property Modeling Data Preparation

Well Log Calculator - Discrete Net-to-Gross (NtG) log example
Calculate a discrete Net-to-Gross log based on Vsh, Porosity and Sw:

1. Create a NtG discrete log template.

2. Use the Calculator to make a NtG discrete

log representing cutoff criteria NtG (1) and
NoNtG (0).

This is an example of how to calculate a discrete Net to Gross log

based on different property cutoffs, such as shale volume, porosity and
water saturation. The output will be a discrete log indicating if there is
a Net interval or Not, according to the codes defined on the NtG
template (as shown in the slide example No NtG=0 and NtG=1).
Property Modeling Property Modeling Data Preparation • 149
 Property Modeling Data Preparation
Continuous NtG log – Upscale log to property & Make synthetic log example
Calculate a continuous Net-to-Gross (indicator) log based on Vsh, Porosity and Sw:

1. Use the Calculator to make a NtG continuous log

representing cutoff criteria. Use the N/G template. 1
2. Open the Scale up well logs process and upscale
the new NtG continuous log using Arithmetic mean.

3. Go to the Wells Settings > Make logs tab > From

property and select the upscaled NtG continuous
log and click the Make logs button.

3
2 3

Net-to-Gross based on Vshale, Porosity and Water saturation.

This is an example of how to calculate a continuous Net to Gross log.
The Net to gross is normally calculated from facies or using cutoffs
based on key logs. The resulting log is a indicator (binary) log showing
Net (1) or No-Net (0), but can be either discrete or continuous in
Petrel depending on which template is chosen for the Calculation.
Here, we chose a Continuous template since we want to upscale and
model NtG in a 3D grid. If it was only used as an indicator log in well
section, a discrete Template would be better.
When the NtG indicator log is Upscaled it will get an average value
(continuous), indicating how much Net to Gross you have as a fraction.
The procedure consists of two main steps:
1. Calculate a NtG log by using different cutoff levels for Vsh,
Porosity and Sw. The result will be a continuous log that
The criteria for the indicates if there is a NtG (1) or not (0) according to the cutoff
cutoff depends on the
property values of each criteria.
reservoir. 2. Scale up the NtG log. The values of the log will be averaged
and resampled into the cells penetrated by the well path,
150 • Property Modeling Data Preparation Property Modeling
 resulting in the NtG value in fraction or percentage for every
cell. Note that the sampling of the upscaled log depends on
the layer thickness.
3. OPTIONAL: Go to the NtG continuous log Settings >
Operations tab to Resample (button) the log.

Property Modeling Data Preparation – Exercises

If you already have a project open, continue on the project you made
and saved in the previous module. If not, use the pre-made Petrel
project stored in Projects folder: Property Modeling [Link].
There are different ways of generating a facies log in Petrel. You can
calculate a lithofacies log using the Well calculator, interactively
interpret facies in a Well section or use Neural Networks in the Train
estimation model process.
Facies calculation
In this exercise, the Well calculator is used to calculate a facies log
based on other logs such as GR, PHI, etc. This should be the first
approach when defining a facies log as it is fast and can provide you
with information for all wells at the same time. To do this, we need to
make a customized template property. Afterwards we can create simple
discrete cut-off logs based on existing logs.
Exercise Steps
1. Open the Windows pane>Module 6 Data Preparation
folder and display the Well Section Facies Interpretation. It
contains panels for GR and PHI.
2. Go to the Templates pane, expand the Discrete property
templates folder>right-click>Insert new property
template. Double-click on this new template (called ‘Untitled 1
1’) and edit the Name and Use icon in the Info tab.
3. Assign the name ‘Simple Litho’ and select the icon used for
Lithologies.
4. On the Colors tab, use code 0 for ‘shale’ and 1 for ‘sand’.
Assign meaningful colors and patterns to each item. Click
OK.

Property Modeling Property Modeling Data Preparation • 151

 5. Open the Calculator by right-clicking on Expl Wells folder.
6. Use the GR log as criteria and choose 70 as the cutoff value to
separate the reservoir interval into sand and shale by typing in
Litho_facies = If(GR<70, 1, 0).
7. Select Simple Litho as the log template and click ENTER.

8. The ‘Litho_facies’ log will be created and attached to the

selected property template. It is stored in the Global well
logs folder. Select the new Litho_facies log to display in the
Well Section Facies Interpretation.

Facies Interactive Interpretation

Facies logs can also be created interactively from scratch, or you can
edit on existing facies logs. Interactive editing of the facies borders can
be necessary to produce a geologically reasonable facies interpretation.
Exercise Steps
1. On the Well Section Function bar, select the Paint discrete
log class button.

152 • Property Modeling Data Preparation Property Modeling

 2. Create a new facies log by clicking the Create new discrete
Log button which is now active. Select Simple Litho as
template for the new log. The new log will be added to Global
well logs folder. Display it on the Well Section Facies
Interpretation window.
3. Click the Paint discrete log class button. Right-click on
the ‘Simple Litho’ track panel for one of the wells in the
Well section to select which facies type to draw, Left-click to
draw the facies and to move facies boundaries.

4. Continue your interactive interpretation for other wells on the

track panel for ‘Simple Litho’.

Neural Networks Classification

The goal of this exercise is to learn how to use the Train estimation
model process and to see the differences of unsupervised and
supervised Neural nets.
Unsupervised Neural net class: Here, the user specifies the number
of classes and the algorithm will separate the data as logically as
Property Modeling Property Modeling Data Preparation • 153
 possible into this number of classes (classification only).
Supervised Neural net class: The user specifies the training data to
be used in the classification or the target for the estimation.
Exercise Steps
1. Go to the Utilities folder in the Processes pane and double-
click on the Train estimation model process.
2. Specify to Create a new model and select Classification.
Select Well logs as Data type.
3. Select wells DW2 and DW6 and the logs GR, PHI and Perm.

4. Click the Correlation analysis button to output the

correlation between the input logs (Gamma, PHI and Perm)

154 • Property Modeling Data Preparation Property Modeling

 5. Under the Settings tab, select Unsupervised and use 2
classes. Leave the other settings as default.

6. Click Apply and look at the Petrel message log to see how
the Relative error is reducing, until the training is stopped by
one of the training criteria (here, there is a 20 iteration limit).

Property Modeling Property Modeling Data Preparation • 155

 7. Petrel has now generated a new Global well log called
‘General discrete (Neural net 1)’. Toggle it on and view it in
the Well Section Facies Interpretation window – compare
it to the ‘Litho_facies’ log. Note that the Neural Net log is
only generated for DW2 and DW6 (input wells to
Classification).
8. Neural net 1 is stored at the base of the Input pane.
9. Make a comparison of which color codes in the Class filter of
the Neural net 1 training set best match the ‘Litho_facies’
Comparing NN to a codes.
standard cutoff/
interpretation, there will be a
bit of deviation in the
‘interpretation’ of the
lithology, but many of the
main features may be picked
up.

156 • Property Modeling Data Preparation Property Modeling

 Notice how the Class 0 can correspond to the Sand facies and
Class 1 to Shale facies.
10. Now, you can also supervise the Neural net classification
process with the ‘Litho_facies’ log: Create new model. Use
same input as last time.
11. In the Settings tab, select Supervised and select the
Litho_facies log from the drop-down list.

12. Click Apply.

13. In Well Section Facies Interpretation, toggle on the new
log ‘Litho_facies (Neural net 2)’ from Global well logs and
compare it to the two existing logs. It should be quite similar to
the log used as the supervised log, and uses the same name.
14. Neural net 2 is stored at the base of the Input pane. Check
the folder as well.

Property Modeling Property Modeling Data Preparation • 157

 Optional: Neural Networks to Make Well Logs
The relationship between PHI, GR and core analysis of Permeability
(Perm log) established in the previous exercise can be used to create
estimated Permeability logs for the wells without Core Permeability.
The Neural net 2 model for Facies will be used to generate a Facies
log for other wells as well.
Exercise Steps
1. Double-click the Make well log process, under
Stratigraphic modeling in the Processes pane.
2. Select the Neural net 2 model (Classification) created in the
previous exercise and drop it into the process by using the blue
arrow.
3. Alternatively, click on Update existing log and select
‘Litho_facies (Neural net 2)’.

4. In the Wells tab, select the wells to create the new facies logs
for as shown in the picture.

5. Click OK and view the results in the Well section window.

Both logs and Neural net can be used later in Facies modeling.
158 • Property Modeling Data Preparation Property Modeling
Create a Water Saturation (SW) Log and Property
In this exercise, the Calculator for Global well logs and properties are
used to calculate a SW log/property based on different methods such
as Archie’s equation, the J-function and by using a macro with
predefined parameters.
A sandstone reservoir is interpreted for the dataset and the SW
equation constants and some other parameters are assumed according
to this interpretation.

Create SW log by using Archie’s equation

Exercise Steps
1. Open the Windows pane>Module 6 Data Preparation
folder and display the well section Sw Calculation. It
contains tracks for GR, RT and PHI_temp, in addition to well
tops (‘Well Tops Volumetrics’) for the OWC (oil-water contact
in both zones A and B). Display also the Perm_temp log
generated in Module 2. Make a visual evaluation to identify
the reservoir intervals.
2. Go to the Expl Wells folder>right-click>Calculator and
enter the equation for the calculation as shown in the figure
below. To use the PHI_temp and RT logs remember to select
them from the Select well log variable window.
3. Select the Sw template from Attach new to template and
Sample MD with>From log>RT.

4. Click ENTER to run the calculation. The new log ‘SW_Archie’

is stored under the Global well logs folder, display it in the
well section and quality check it.
Property Modeling Property Modeling Data Preparation • 159
 5. Open the Settings for the ‘SW_Archie’ log and check the
Statistics and Histogram tabs to make sure that the SW has
the correct range, between 0-1.
6. If the ‘SW_Archie’ log does not have the correct range or
undefined values, it can be fixed by doing a simple calculation
as shown in the next figure:

7. Visualize the changes on the well section.

8. Make sure to color fill the curve by using the Create/edit
curve fill button.

Optional: Create Sw Property by Using a Macro in the

Properties Calculator
To generate a SW property in 3D, a macro file can be loaded. This
makes it easier to keep track of the various statements as the Petrel
calculator only has one statement line. If you use a file, you can have
multiple lines. In the exercise below, a file is loaded using 4 lines,
creating a Temporary file for the existing 3D properties: These
properties are Permeability (Permi) and Porosity (Poro) in relation to an
Above fluid contact property (AboveContact) looking at OWC in two
zones (A and B).
Exercise Steps
1. Continue on the same well section Sw Calculation. Go to the
Models pane> Exercise Model> Properties> right-click>
Calculator and select on the option From file.
2. Click on the Browse button and select the macro file to
be used. The macro file is stored in the directory: Property
Modeling2010 >Data>sw_formula.mac.
3. Select the Sw template from Attach new to template.
4. Click the button. Notice that all of the formulas will
appear in the history window at the top. Two new properties
will be created Temp and Sw.

160 • Property Modeling Data Preparation Property Modeling

 5. The new property Sw is stored under the Properties folder,
display it in the well section and quality check. Make sure it is
in a separate track and color filled. Compare with the SW_
Archie and SW_Advanced logs.
6. Also, have a brief look at the Sw, Poro, Permi and
AboveContact properties in a 3D window using I, J and K
filters.

Property Modeling Property Modeling Data Preparation • 161

 Create a Net-to-Gross (NtG) Log
In this exercise, the log Calculator is used to make a NtG log using the
Vshale (Volume of shale), Porosity and SW (Water Saturation) logs.
Based on different cutoffs, specifically for this model, assume that a
good reservoir zone has: Vshale<= 0.45, Porosity>=0.15 and Sw<=0.8.
Two methods will be used to calculate the NtG log. The output will be a
discrete log indicating if there is a reservoir zone or not, and a
continuous log showing different values of reservoir thickness.
Create a Continuous NtG log (well log calculator)
Exercise Steps
1. Open the Windows pane>Module 6 Data Preparation
Remember that the folder and display the well section NtG Calculation. It
different logs used for the contains panels for Vsh and Poro. Display the Sw_Archie log
calculation are selected from
Select well log variable you just created in the previous exercise. Make a visual
window. Select the N/G evaluation to identify the reservoir intervals and to verify the
template from Attach new average cutoffs that have been proposed for this exercise. Feel
to template>Petrophysical
templates. free to define your own cutoffs for the calculation.
2. Go to the Global well logs>right-click>Calculator and enter
the equation for the calculation as shown in the figure below.
Make sure you select the continuous Net/Gross template.

3. Click ENTER to run the calculation. The new log ‘NtG_cont’ is

stored under the Global well logs folder, check on the log box
to display it in the well section and quality check.
4. In the Models pane, make sure that the Exercise Model grid
is selected.
162 • Property Modeling Data Preparation Property Modeling
 5. Open the Scale up well logs process>Create new property
and select ‘NtG_cont’ log. Leave all the settings as default
and click OK.
6. The new upscaled ‘NtG_cont’ log is stored in the Properties
folder. Display it in the Well section and compare with the
‘NtG_cont’ well log.
7. In the Input pane, go to the Wells folder>Settings>Make
logs tab>From property, select the NtG_cont (U) property
and click on the Make logs button.
8. The new ‘NtG_cont’ log is stored in the Global well logs
folder. Rename it to ‘NtG_cont_synt’ and check on the log to
display it on the well section.
9. Optionally, go to the Settings of the ‘NtG_cont_synt’ >
Operations tab and click on the Resample In the well section,
use the Create/edit curve
button using 1 m sampling. Now, visualize the result on the
well section. fill icon together with
the Well sections
template Settings>
Template objects> Track>
Curve filling to define the
desired color fill for the logs.

Property Modeling Property Modeling Data Preparation • 163

 Optional: Create a Discrete NtG Log (well log calculator)
Exercise Steps
1. Continue on the same well section NtG Calculation.
2. Go to the Templates pane, expand the Discrete property
templates folder>right-click>Insert new property
template. Double-click on this new template (called ‘Untitled 1
1’) and go to the Info tab.
3. Assign the Name and Legend label ‘NtG’ and select the Use
icon as General discrete.
4. On the Colors tab, use code 0 for ‘NoNtG’ and 1 for ‘NtG’.
Assign meaningful colors and patterns to each item. Click
OK.

5. Go to the Expl Wells folder>right-click>Calculator and

enter the equation for the calculation as shown in the figure
below. To use the Vsh, Poro and SW_Archie logs, remember
to select them from the Select well log variable window.
6. Select the new discrete NtG template.

7. Click ENTER to run the calculation. The new log ‘NtG_disc’ is

stored under the Global well logs folder, select the log box to
be displayed on the well section and quality check it.
164 • Property Modeling Data Preparation Property Modeling
Summary
Facies log preparation using the interactive drawing of a facies log
based on petrophysical logs, as well as working with the well log
calculator has been covered in this module. In addition, combined
calculations using logs has shown us how to create logs ready for input
to a 3D grid.
Furthermore, the basic concept of Neural Network method has been
demonstrated in this module. The ‘Train estimation model’ (as the
process is called in Petrel) can be divided into classification techniques
for discrete data, and estimation techniques for continuous data. It can
be applied to different types of input data, such as well logs, seismic
cubes, 3D properties and point data or surfaces. Analysis tools
(Correlation) help to filter relevant input information, and supervised
models incorporate a-priori information as training images to optimize
the output result.
Finally, the use of Well log and Properties Calculators for logs and
properties such as Water Saturation and Net-to-Gross, based on
different methods and petrophysical logs cutoffs have been shown.

Property Modeling Property Modeling Data Preparation • 165

Module 7 - Scale Up Well
Logs

How to prepare the inputs for Property modeling through the Scale up
well logs process will be presented in this module. This process
assigns the log property values (both discrete and continuous) to the
cells penetrated by the wells in the model. We will also focus on how
to quality control the result of this process.

Prerequisites
To successfully complete this module, the user must have knowledge of
the following:
• Petrel Introduction
• Well Correlation

Learning Objectives
The purpose of this topic is to prepare the well logs as an input for
Property modeling. At the completion of this training, you will be able
to:
• Understand and use discrete and continuous logs for property
modeling
• Utilize different settings for scaled up well logs
• Look at various considerations when performing upscaling
related to wells and 3D grid
• Learn how to quality control the upscaling process using well
sections and histograms

Property Modeling Scale Up Well Logs • 167

 Lesson

Scale Up Well Logs

Process

Cells along the well path

Log values assigned to cells

Up-scaled logs used to fill up the

3D grid with properties

Scale up well logs - Principle

Scale up well logs is the process of assigning values to the cells in
the 3D grid that is penetrated by the wells (the pillars define the
horizontal size and the layering defines the thickness of each grid cell).
Since each cell can only hold one value, the well logs must be
averaged; for example upscaled. The goal is to use the well information
as input for the Property modeling process for the distribution of
property values between the wells.
It is important to be aware that the upscaled cells will be part of the
property and not a separate item. Consequently, the value of the cells
along the well path will be the same in the entire 3D property as for the
upscaled cells alone.
The principle is as follows:
1. Assign property values to the cells along the well path by
upscaling the log values.
2. Based on information from the upscaled logs, fill the entire 3D
grid between the wells.

168 • Scale Up Well Logs Property Modeling

 Scale Up Well Logs
What to consider when upscaling well logs
Discrete Logs Continuous Logs
Integers (0,1,2, etc), such as for facies logs. Real numbers (decimals), such as for porosity logs.

1. Select the average method (Most of). 1. If a discrete log (e.g. facies) is available, use
2. Define how the grid cells penetrated bias to this if necessary.
by the well are chosen. 2. View log histogram to examine property
3. If applicable,use facies weighting and distribution
view the changes in proportions in the 3. According to the type of log property and
log histogram. distribution select the average method.

Scale Up Well Logs

Discrete Logs – Facies/Lithology
The Average method is normally Most of
(the value which is most represented in the
cell will be used).

In some situations it may be necessary to

use the weighting option. Toggle the Use
facies weighting and fill in Weighted tab.
Adjust the weights and see proportion
changes in the Histogram.

Property Modeling Scale Up Well Logs • 169

 Scale up Discrete well logs
For discrete logs, the common Averaging method is the Most of, where
the majority wins. It assigns the most frequently occurring log values to
each cell. If there is 49 % Shale and 51 % Sand, the cell will be
populated by Sand only (which there is most of).
There are options for treating logs as points or as lines. If the log is
treated as lines, the values that are actually outside the cell can
contribute to the averaging because an interpolation has been done
between points. If the log is treated as points, only the log points
themselves will be used as input for the averaging.
For interactively created facies logs, use line option since only the
facies boundaries are recorded.

Averaging methods
Most of - Will use the value which is most represented in the cell.
Median - Will sort the input values and select the center value. For
example if there are 7 input values, these are sorted by magnitude and
then entry number 4 in the sequence is selected.
Minimum / Maximum - Samples the min/max value of the well log
for the cell.
Mid Point Pick – Will pick the log value where the well is halfway
through the cell. This is essentially a random choice and is therefore
more likely to give a property with the same distribution of values as
the original well log data.
Random Pick – Picks a log point at random from anywhere within the
cell. This random option avoids the smoothing tendency of other
methods and is therefore more likely to give a property with the same
distribution of values as the original well log data.

Considerations
The Most of method is suitable in cases where a right sizing of the grid
layering has been performed (variogram analysis) and the heterogeneity
and spatial distribution of facies are captured. However, sometimes
there may be thin layers of shale or anhydrite covering a large area –
170 • Scale Up Well Logs Property Modeling
this may act as a barrier and are important in reservoir modeling as they
may inhibit flow. If they are thin (i.e. always less than 50% of a cell),
they will never be recorded. The Layering must be redone to capture
these streaks of important facies.
Random Pick – This method could be an alternative in uncertainty,
when cells have similar proportions; if Weighting is used it will always
favours one facies, while a Random pick will alternate between
almost similar proportions (however if a very small fraction facies is
represented in a cell, it may ALSO be sampled).

Weighting
The Weighted tab can be used when discrete logs are upscaled. This
tab is enabled when choosing Use facies weighting. With this
option, it is possible to change the weight of a particular facies, or to
adjust the proportion of the upscaled log to the raw log. By default, all
of the facies will have a value of 1 on the Weighted table, the values
are relative, so they can be assigned as fractions.
To see the effect while working with the Weighted tab:
1. First upscale the discrete log Weighting will
always favor one or more of
2. Open Settings>Histogram, apply filters if needed your facies manually; for
3. Define the weight value on the Weighted tab table example you should be SURE
that these facies are the
4. Click Apply and visualize the changes for the facies fraction in ‘correct’ ones, or somehow
the Histogram/Well section. Redo if necessary. under-represented in the
upscaling. If unsure, try
redoing the Layering to get a
This method may be useful when modeling reservoirs with production more correct/representative
on layers with fine lamination. upscaling.

Property Modeling Scale Up Well Logs • 171

 Scale Up Well Logs
Continuous Logs – Petrophysical Properties

Use bias to facies if the continuous value

should get an average value closely
representing the discrete (Most of) value

Select the Average method. It is normally

Arithmetic for porosity (averages all the
values in the cell equally).

You can specify how to treat the log

values into the cells As Lines or Points.
As well, the method to define how the grid
cells penetrated by the wells are chosen.

You can specify that there should be a

Min. number of points in cell passing
through for it to be populated with a value.

Scale up Continuous well logs

The purpose of upscaling continuous well logs is to get an average log
value for each cell penetrated by the well.
The most used averaging methods are Arithmetic, Geometric and
Harmonic. As a general rule, arithmetic averaging will always give
higher values than geometric averaging, which will give higher values
than harmonic averaging.
Averaging methods
Arithmetic averaging - is usually the best method for porosity.
Arithmetic/Harmonic and Geometric - is common for different types
of Permeability.
RMS (Root Mean Squared) - Will provide a strong bias towards high
values (positive).
Median - Will sort the input values and select the center value, for
example if there are 7 input values, these are sorted by magnitude and
then entry number 4 in the sequence is selected.
172 • Scale Up Well Logs Property Modeling
Minimum /Maximum - Samples the min/max value of the well log for
the cell.
Mid Point Pick – Will pick the log value where the well is halfway
through the cell. This is essentially a random choice and is therefore
more likely to give a property with the same distribution of values as
the original well log data.
Random Pick – Picks a log point at random from anywhere within the
cell. This random option avoids the smoothing tendency of other
methods and is more likely to give a property with the same distribution
of values as the original well log data.

Scale Up Well Logs

The concept of Biasing to a Discrete Log
Continuous logs can be biased; i.e. use averaging only in the portion of
specified Facies, Net-to-gross or other discrete logs.
Upscaled facies Raw facies Raw porosity Upscaled porosity

Sand

Bias to a discrete log

Continuous logs, such as porosity, can be (and usually should be) biased
to a discrete log, such as the facies log. The purpose of this is to keep
cleaner statistics, and make sure the petrophysical values match better
the identified facies (decided by the majority rule of the “Most of”
method).

Property Modeling Scale Up Well Logs • 173

 Example:
Based on several logs, a raw facies log has been generated. When the
facies log is upscaled to fit the 3D grid layers by using the Most of
method, some of the cells will be given a sand value, even if the cell
has porosity values that correspond to both shale and sand. If the
porosity logs were to be upscaled without using a bias, then an
averaging would be done on the raw porosity log to produce an
intermediate result that does not represent either sand or shale. By
using the facies log as a bias, only the porosity values that
correspond to the most frequently occurring facies in the current cell
will be used as input for the averaging, and as such, create a porosity
value that represents the upscaled facies value only.
It is assumed that the upscaled cells will sometimes be assigned the
values representing the shale, even if there are sand values in the raw
logs. Other cells will be assigned the values representing sand, even if
there are shale values in the raw logs. Generally, the mistake of
creating too much sand in some areas, and too much shale in other
areas, is evened out.

SW (water saturation) - upscaling logs

During upscaling, SW can be biased to a Reservoir log, whether it is an
Reservoir Rock Type (RRT) or Facies log. Alternatively, a Discrete NtG
log (see next two slides) can be used as bias.

NtG (Net/Gross) - upscaling logs and modeling

Upscale a binary (1 and 0) NtG discrete log if you need to use it as bias
(See Module 6).
Upscale a binary (1 and 0) N/G continuous log if you need a continuous
N/G distribution in Upscaled position to be used as basis for 3D
modeling.
Then model NtG conditioned to Facies, also where there are sections of
non-reservoir.

174 • Scale Up Well Logs Property Modeling

 Scale Up Well Logs
Porosity / SW / NtG (Arithmetic Mean)

Arithmetic mean is typically used for properties such as Porosity,

Water saturation (SW)* and Net to gross (N/G) because these are
additive variables.

1 n
Arithmetic Mean xa = ∑ xi
n i =1

10,00
8,00
6,00
4,00
2,00
0,00
0,01 0,11 0,21 0,31

* See next slide

Scale Up Well Logs

SW (Arithmetic Mean vs. Median/Mid-point-pick)

Water saturation (SW) can have a skewed shape and should consequently be
upscaled using an alternative method.

Arithmetic Median

Property Modeling Scale Up Well Logs • 175

 SW (water saturation)
Due to the skewed nature of most SW distributions, arithmetic
upscaling will be too optimistic (in terms of SW); therefore too
pessimistic (in terms of Oil-in-place).
Using a “Mid-point-pick” or “Median” method will follow the
distribution better. Also, the “Random pick” can be used, but with
caution.

Scale Up Well Logs

Permeability Logs (Arithmetic Mean)
Permeability governs the flow of oil/gas/water.
Investigation has shown that Arithmetic Mean may give a
too optimistic value for permeability.

When to use Arithmetic Mean:

For homogeneous layers and flow parallel to the bedding.

1 n
Arithmetic Mean xa = ∑ xi
n i =1
10,00
8,00
6,00
4,00
2,00
0,00
0,01 0,11 0,21 0,31

176 • Scale Up Well Logs Property Modeling

 Scale Up Well Logs
Permeability Logs (Harmonic Mean)

When to use Harmonic Mean:

For homogeneous layers and dominant vertical flow
to the bedding. Harmonic mean gives the effective
vertical permeability if the reservoir is layered with
constant permeability in each layer. It works well with
log normal distributions. It is sensitive to lower values.

n
Harmonic Mean xh = n
1
∑x
i =1 i
0,14
0,12
0,10
0,08
0,06
0,04
0,02
0,00
25 75 125 175 225 275 325 375

Scale Up Well Logs

Permeability Logs (Geometric Mean)

When to use Geometric Mean:

For average permeability of random heterogeneous
layers. It is normally a good estimate for permeability if
there is no spatial correlation and the input data is
log normally distributed. It is sensitive to lower values.

Geometric Mean xg = n ∏x
i =1
i

0,07
0,06
0,05
0,04
0,03
0,02
0,01
0,00
25 275 525 775 1025 1275 1525 1775

Property Modeling Scale Up Well Logs • 177

 Scale Up Well Logs
3D grid considerations – Cell size (I, J, K)

If vertical wells are upscaled, Layering is important to get the correct resolution.
However, if lateral wells are upscaled, the I,J increment is important for resolution

X,Y (I,J)

Z (K) Overburden

Channel sand

Cell adjustment for deviated/horizontal wells Pillar grid process

Cell adjustment for vertical wells Layering process

Cells size in I,J,K direction

In the Horizontal part of a 3D grid, the resolution is generally much
coarser as it is decided by the I, J cell size set in the Pillar Grid (seen
along MD in the Well section). Laterally you would expect a much
coarser resolution of properties due to the nature of time and
deposition.
In the Vertical part, the averaging of the logs are more critical as the
expected variation / heterogeneity is much higher than laterally (not
always), so then the Layering is important in deciding the optimal cell
height to capture that variability.
Example:
An analogy to how I,J varies with respect to K direction is seen in the
way variograms are set up. For example, in a clastic sandstone reservoir
with channel deposits, the lateral variation along channels can be quite
uniform, while the vertical variation is much higher and require a much
finer resolution. In small reservoirs with uniform deposits, like some
178 • Scale Up Well Logs Property Modeling
carbonates or tabular sandstone deposits, the vertical variation may not
be very different from the horizontal. But overall, deposits in the I, J
direction along one K-layer is considered one time-line, so the lateral
variation is generally much higher than the vertical.
Some notes of warning regarding horizontal wells and upscaling is that
due to a heavier averaging (larger cell size) there will be a volume
support effect, and extreme values will have a higher probability of
not being captured.

Scale Up Well Logs

3D grid considerations – Well type and spacing

When upscaling, the well spacing has to be considered w.r.t Cell size; if well spacing is
less than I,J – an averaging of two well logs is performed in one cell.

Method Description
Simple All cells penetrated by the well
trajectory are included
Through cell The well trajectory must penetrate
two opposite cell walls
Neighbor cell Cells in the same cell layer are
averaged.

Averaging Logs from Multiple Wells

The value that is put in a cell along a well trajectory is dependant upon
the log sampling, how many wells penetrate the same cell and what
type of cell-pick is selected in the upscaling process. The actual cells
that are picked for averaging is decided by the cell selection method
(Simple, Through cell or Neighbor cell).
If a cell is penetrated by more than one vertical well, the value is an
average of the logs in the wells.
Property Modeling Scale Up Well Logs • 179
 If the wells are deviated, then also the method used is important for the
sampling and the actual value (Simple and Neighbor cell will select
more cells, but will have different averaging for neighboring cells in the
same K-layer). The Through cell method is more accurate but will
probably sample too few cells.

Upscaling Methods
The input values from the raw logs to be used for averaging, can be
selected in different ways:
Simple – In this case all the cells penetrated by the well will be
assigned a value, even if just a tiny corner of a cell is penetrated by the
well.
Through cell – In order for a cell to be assigned a value, two opposite
cell walls must be penetrated by the well path. This method can be
used to ensure that only a tiny part of the well path does not contribute
to giving an entire cell a value.
Neighbor cells – All cells penetrated by the well path will be
assigned a value (as for the simple method). However, cells next to
each other in the same layer will be averaged. This is usually a good
method to use to avoid a vertical barrier becoming a horizontal barrier.
See the Online Help for a more detailed explanation.

180 • Scale Up Well Logs Property Modeling

 Scale Up Well Logs
3D grid considerations – Well selection
Based on analysis of well spacing or analysis of representative parts of a grid, a
selection of wells can be upscaled. You can use Saved Searches for this.

Once the Saved search is inserted into the Upscaling

process, it knows which filtered wells to use.

Saved searches
When upscaling wells, it is common to select all wells, as there are
often few wells to start with and we need all the information we can
get. However, if many wells are present and some wells are not
representative of what you will upscale and eventually model in 3D,
these wells should not be upscaled.
An easy way of creating a selection of wells is to use Saved
searches. You could, for example, search and use wells from a
predefined list, design a polygon (see slide) around the area you want to
use the wells in, etc.
How to make a Saved search:
1. In the Wells>Saved searches folder>right-click, select
New search.
Property Modeling Scale Up Well Logs • 181
 2. In the Settings>Search criteria, select Filter type from the
drop-down list. In the Info tab, rename it and click OK.
3. Go to the Scale up well logs process>Wells, select the Use
saved search check box and drop it into the process by
selecting the search and click the blue button.

Scale Up Well Logs

Upscaled Logs Quality Control – Well section
QC the upscaled cells with the original log
in a Well section:

Press the Show result in well section or set

up a Well Section display on your own

If setting up a Well Section; toggle raw logs

from Global well logs and upscaled
properties from Properties folder.

View the logs and upscaled cells of interest;

check for inconsistencies. Redo Upscaling if
necessary.

Scale up well logs – Quality Control

Before petrophysical modeling is performed, the upscaled well logs
should be quality controlled. There are several ways of quality checking
the upscaled well logs and the generated property models. In Petrel,
some tools for quality control can be:
• Histogram or the Statistics tab (settings for the selected property) to
review the univariate statistics of the data
• Crossplot to review bivariate statistics, like property correlation
• Variograms to review the spatial statistics of data
182 • Scale Up Well Logs Property Modeling
Quality check
We will quality check the result of the upscaling of well logs by using
histogram analysis and displaying the result in a well section.
When you upscale the logs, it is necessary that the important units are
captured. This means that thin, but extensive, lateral barriers must be
incorporated in the model, and also in the thick flow units. However,
you do not want to have a too fine-scaled geological grid either,
because that will slow down processing time and inhibit visualization
of results. Therefore, it is important that the thickness of the layers
have been set with respect to the important flow units/barriers/baffles.
After the well logs have been upscaled, a QC needs to be performed to
see if the flow units and the barriers/baffles have been captured. If the
layering is too thick, you may have lost too much information and you
will have to adjust the thickness of the layers in your model.

Scale Up Well Logs

Upscaled Logs Quality Control – Histogram and Filter
Channel Porosity
QC the upscaled cells with the original log
in a Histogram. Filter on facies type:

Lobe Porosity

Property Modeling Scale Up Well Logs • 183

 Filters
When performing the quality control, you should combine different
methods, such as displaying both the input data and the result in a Well
Section and Histogram; using different filters.
To visualize in a Histogram, some filters can be applied for zones, facies
or petrophysical properties for example.
Filters can be generated and are subsequently stored in the Properties
folder>Settings>Filter tab or in the Input pane>Filter folder as 1D
filter or logical filters.

Scale Up Well Logs – Exercises

To get values in the 3D grid along the well bore, the well logs must be
upscaled (blocked). This is done by averaging the well log values within
each cell using different averaging methods (typically arithmetic,
geometric or harmonic for continuous well logs and “most of” for
discrete well logs).
This exercise has a pre-made Petrel project stored in Projects folder:
Property Modeling 2010_M7&[Link]. If you did the last exercise,
please use the same project.
Scale up discrete well logs
Exercise Steps
1. Go to the Windows pane and from the Module 7 Scale up
well logs folder, open the Well section LOG-UPSCALING
QC. This section contains GR, Facies, PHI, SW and Perm logs.
2. Make sure the Exercise Model grid is active. Open the Scale
up well logs process found under Property modeling in the
Processes pane.
3. Select Create new property. Scale up the Facies log with
the Most of average method and click Apply (do not click OK
yet). The new upscaled Facies (U) log is stored into the
Exercise Model (3D Grid)>Properties folder, display it on
the Well section.
4. In the Scale up well logs process, select Create new
property and scale up the Facies log with the Most of
average method. Select the option Use facies weighting and
go to the Weighted tab. Deselect Equal for all zones.
184 • Scale Up Well Logs Property Modeling
 5. Set up the weight values for the different facies as shown in
the picture and click Apply.

6. Go to the Facies (U) Settings>Info tab and rename it as

‘Facies Weighting’ and display it in the Well section.
7. Compare both Facies (U) and Facies Weighting (U) with the
original log. Visualize the changes on the Well section and
Histogram for the different zones.
8. Open the Settings> Histogram tab for both upscaled facies.
Only display Upscaled cells and Well logs. Select the Filter
zone option and visualize the facies histogram for both zones.

Property Modeling Scale Up Well Logs • 185

 Scale up continuous well logs
Exercise Steps
1. Go to the Global well logs folder and open the Settings >
Histogram tab for the porosity (‘PHI’ log) and analyze the
histogram distribution shape. The lowest values can be filtered
by typing 0.01 as Min value and use the Refresh button to
apply the changes as shown in the figure below. It will help you
determine which type of distribution the porosity has and
consequently to select the optimal Averaging method for
upscaling.

186 • Scale Up Well Logs Property Modeling

 2. In the Scale up well logs process, select Create new
property again and scale up the Porosity log (‘PHI’), use
‘Facies (U)’ as bias log. The Average method for PHI should
be Arithmetic according to the pre-analysis in the step 1. Click
Apply.
3. Use the same Well section of the discrete logs to display the
PHI log and PHI (U) property.
4. Open the Settings > Histogram tab for the PHI (U) property
and display only Upscaled cells and Well logs to visualize
the result. In the Scale up well logs process, change the bias
log to ‘Facies Weighting’ and visualize the changes, also in
the Well section.
5. Go to the Settings > Statistics tab for PHI (U) and check the
Max. value for Upscaled cells and Well logs. In the Scale up
well logs process, select the option Treat log As points and
notice the changes in the Statistics/Histogram tabs and in
the Well section.

Property Modeling Scale Up Well Logs • 187

 6. Optionally, select the option Min. number of point in cell,
enter values from 3 to 10 and visualize the changes in the
Histogram tab and Well section.
7. Finally, revert back to the initial upscaling settings for PHI,
using bias to Facies (U) and Treat log as lines.
8. Also, scale up the Water Saturation log (‘SW_Archie’).
Upscale it using the average method Arithmetic and click
Apply.
9. In addition, apply the same procedure for the Permeability
(‘Perm’ log) and analyze the histogram distribution shape. The
Perm_temp log can be used to help you in the analysis as it
has been edited to cut away all zero values, as shown in the
figure below.

10. Create new property and scale up Permeability (‘Perm’ log).

Use Facies (U) as bias log and use Arithmetic averaging
according to your pre-analysis of the distribution. Click Apply.
Aditionally, you can test the Harmonic and Geometric
methods as well.
188 • Scale Up Well Logs Property Modeling
 11. Try the different settings until you get the desired result, then
display it in the Well section, check the Statistics and
Histogram for the upscaled Permeability (Perm (U)).

Optional: Quality Control of upscaled logs

It is important to check that you were able to capture the important
features from the raw well logs when you upscaled the values into the
3D grid cells. The following exercises will use the Well section window
for visually displaying the raw well logs together with the upscaled
logs.
Exercise Steps
1. Before you exit the Scale up well logs process dialog, click
the Show result in well section button. Petrel will
automatically create a display of the raw and upscaled logs in
a Well Section window. Do a visual QC of the upscaling of the
discrete logs and the continuous logs.
2. Display the Upscaled properties in a 3D window.
3. For upscaled Facies (U) properties, check the Settings>
Histogram tab and compare the facies proportion for each
facies between the raw log and the upscaled log .
Property Modeling Scale Up Well Logs • 189
 Notice the difference caused by the Scale up well logs
process.
4. For upscaled petrophysical logs, check both the Statistics tab
and the Histogram tab. Notice from the histogram that the
extreme low values and extreme high values in the raw log will
be lost during the Scale up well logs process.
5. Finally, do a clean up by putting the new properties into a new
‘Upscaled’ folder in the Properties folder.

Comments
• Another option is to resample the upscaled logs back to wells
as synthetic logs: Wells (right-click) > Settings > Make logs
> From property. Select the desired properties and press
Make logs.
• All the synthetic logs created will be added to the Global well
You may find a logs folder. Visualize and compare the raw log and synthetic
problem with the upscaled logs in a 3D window to see how well the variation is captured
Perm log. Compare the
different averaging methods
during the Scale up well logs process.
for the Perm log. In this • There are always some features that are lost during the Scale
case Arithmetic method is up well logs process. You can reduce the 3D grid cell
recommended. Also try the
Maximum and Random pick thickness to minimize the loss. The decision needs to be made
methods. to balance the cell number and accuracy.

190 • Scale Up Well Logs Property Modeling

Summary
How to scale up well logs into the cells penetrated by the well path and
how to quality check by displaying the results in a Well section and 3D
window, have been covered in this module. You have also used the
Statistics and Histogram tabs to do a statistical check of the raw and
upscaled log values. This is a pre-process to subsequently prepare the
inputs for the Facies and Petrophysical Modeling.

Property Modeling Scale Up Well Logs • 191

Module 8 - 3D Grid Quality
Control Tools

This module offers an overview of the statistical tools for quality

checking of the input data before performing property modeling. How to
use vertical variogram to quality check the 3D grid layer resolution is
also covered in this module.

Prerequisites
To successfully complete this module, the user must have knowledge of
the following:
• Familiarity with Geostatistics Fundamentals
• Petrel Introduction

Learning Objectives
The purpose of this module is to show you how to perform quality
control of the 3D grid layer resolution before the Property modeling is
performed. At the completion of this training, you will be able to:
• Perform a quality check of the grid layer resolution in a vertical
variogram to make sure that proper grid resolution is present
for the properties.

Property Modeling 3D Grid Quality Control Tools • 193

 Lesson

3D Grid Quality Control Tools

Overview

Univariate Statistics
 General statistics
 Histogram analysis
Bivariate Statistics
 Crossplot
 Regression line/Correlation
Spatial Statistics
 Vertical variogram analysis

Quality Check Input Data and 3D Grid Resolution

There are several ways of quality checking (QC) the generated property
models. In Petrel, there are tools for displaying the data as a histogram
or in a cross plot.
Before Petrophysical modeling is performed, the grid layer resolution
should be quality controlled. In the exercises in this module, we will do
so by performing vertical variogram analysis.
The goal is to determine if a proper grid resolution is present for a given
property. You can save time by performing this analysis prior to
populating properties (as opposed to wasting time via trial and error
attempts at property modeling using an incorrect grid resolution).

194 • 3D Grid Quality Control Tools Property Modeling

 3D Grid Quality Control Tools
Univariate Analysis - Discrete properties
Univariate Statistics (Discrete properties) describe a particular data set with
corresponding codes for one discrete variable.

0
Histogram
1
2
3
Discrete statistics

Univariate Statistics (Discrete properties)

Histogram
The distribution of a discrete property can be displayed in a set of
histograms showing the distributions for both the upscaled cells and
the raw well log data that served as input to the Scale up well logs
process. The comparison of these two distributions (upscaled vs.
original raw logs) can reveal if the upscaling has dramatically changed
the distribution. For discrete data, a histogram column only represents,
for example, a facies type (by code, as given in the template).
Statistics
The Statistics tab for discrete properties in Petrel is not very
informative in terms of mean, max, min and [Link] values as it only
looks at the discrete code numbers.
However, the Discrete statistics (Disc. Stat.) tab provides a better
overview where Min, Max, Mean and [Link] is for interval heights
while the number in brackets gives the number of cells.

Property Modeling 3D Grid Quality Control Tools • 195

 3D Grid Quality Control Tools
Univariate Analysis - Continuous properties
Univariate Statistics (Continuous properties) describe a particular data set with a
series of measurements unique to a single set of data assumed to represent values for
one variable.

Histogram

Statistics

Univariate Statistics (Continuous properties)

Histogram
The distribution of a property can be displayed in a set of histograms
showing the distributions for both the upscaled cells and the raw well
log data that served as input to the Upscale well logs process.
Additionally, the histogram is a tool to check the distribution of the
given dataset. A normal distribution is required by most of the
algorithms. Bi- or multi modal distribution is an indication of secondary
processes in the reservoir.
The comparison of these two distributions (upscaled vs. original raw
logs) can reveal if the upscaling has dramatically changed the
continuous distribution. However, this semi-quantitative analysis alone
may not be conclusive in terms of helping you determine the proper grid
resolution. At best, the histogram analysis alone may require some
degree of trial and error before the proper resolution is discovered. You
can also identify a problem, but not necessarily locate it. In other
196 • 3D Grid Quality Control Tools Property Modeling
words, you may see gross problems by comparing the histograms, but
you will normally not know what the final grid resolution should be from
this analysis.
Statistics
The Statistics tab for a given continuous property provides a good
overview of Max and Min values, as well as Mean and [Link].

3D Grid Quality Control Tools

Univariate Analysis - Filters (Setup)
Filter Setup: To get a better picture of the univariate statistics, it is useful to filter out
discrete values to narrow down/inspect statistics related [Link] certain facies types.

Property Filters
The property filter is a tool for removing cells of given values from the
display and the related statistics representation according to a given
filter criteria. It enables you to display the property data in different
histograms that each represent different facies types, for example, the
porosity distribution within the channel facies.
Hint: When you use the filter for univariate statistics, remember to
choose the option Upscaled: As normal cells (this will also filter the
upscaled cells). If the option Always include is selected, the filter in
the Statistics and Histogram tabs does not filter the upscaled cells.

Property Modeling 3D Grid Quality Control Tools • 197

 3D Grid Quality Control Tools
Univariate Analysis - Filters (Applied result)
Applied Filter: A filter can be set up on a discrete property but is automatically applied
on other properties (e.g. the Porosity values can be filtered for each facies type). For
applying the filter in the Histogram or Statistics tab activate the filter icon.

Channel Levee
Porosity Porosity

Lobe Shale
Porosity Porosity
Porosity Distribution
for all Facies types

3D Grid Quality Control Tools

Bivariate Analysis

Bivariate Statistics explores the

relationship between two variables.

Correlation Analysis testing if a

significant coincidence between two
variables exist.

Regression Analysis quantifies an

existing correlation (here by a linear
formula).

198 • 3D Grid Quality Control Tools Property Modeling

Why is Correlation important?
To establish whether a correlation between two datasets or variables
(like logs or properties) exist or not is important in reservoir modeling as
it enables the use of a secondary property in the modeling. It is also
important in the QC of the layering of a 3D grid, as it can capture
deviations from an expected correlation resulting from bad grid
definition or wrong upscaling. Typically, high resolution data from logs
can be correlated to lower resolution properties of the same type which
are resampled in the 3D grid. For example, too dense layering can cause
noise in the crossplot points.
Correlation analysis
Measures the strength and direction of a linear relationship between
two variables.
Regression analysis
The dependent variable in the regression equation is modeled as a
function of the independent variables, for a best fit. It can be used for
prediction, inference and modeling of causal relationships. It quantifies
an existing correlation which generally can be both linear and non-
linear.

3D Grid Quality Control Tools

Bivariate Analysis – Correlation
Crossplot: The Function window is used for exploring 2 variables. Select the X-value first,
then the Y-value. Alternatively a third Z-value can be used for coloring the sample points.
Calculate the correlation coefficinet by using the Make linear function icon.

Property Modeling 3D Grid Quality Control Tools • 199

 Crossplot
A crossplot can be set up in Petrel using the Function window. The
first property values selected will be displayed on the X-axis, the
second values on the Y-axis and, alternatively, a third value can be used
for coloring the sample points at the same cell location. If permeability
or any other logarithmic values are displayed, a log scale on either axis
can be set.
When you create a
new correlation line, the By using the Show/hide autolegend button, the coloring of the third
Correlation coefficient is variable can be displayed in the Function window. This is useful for
ONLY displayed in the dialog fast identification of correlation/no correlation domains.
window, and is not recorded
in the settings of the line The property filters are also active here and can be used to filter certain
(generated and stored as a
separate object in the Input values. A new correlation line can be generated based on the filtered
pane). values.

3D Grid Quality Control Tools

Bivariate Analysis - Regression

Correlation line: Input data is used to generate a regression line using a linear
function. If a filter is on, the correlation line is automatically calculated from the filtered
values, and can be edited in the Function window or Function tab.

Edit point

Edit line

Original linear function used for

creating the correlation line

Edit Correlation Line

The correlation line generated based on all or only filtered input data is
stored in the Input pane. The original linear function used for
200 • 3D Grid Quality Control Tools Property Modeling
generating the line is stored in the Info tab of the object.
You can edit the line, using a point selector or a line selector. New If the original
points can also be added to generate a non-linear line. correlation line is edited, the
linear function used to create
the line is not updated in the
3D Grid Quality Control Tools Info tab.

Spatial Analysis – Vertical Variogram

Spatial Statistics depend on the relative position of the data to each other
Analyze raw log data to determine resolution requirements in the
vertical direction
Compute experimental variogram in the vertical direction
Fit spherical variogram model to the experimental variogram

Justification
- Vertical sampling of raw log data is usually very dense
- Vertical range indicates the distance within which data are correlated

Layer resolution should be at or below ½ of the vertical range

Variogram Analysis
Analysis of variograms can be used to determine the required grid
resolution. To aid you in selecting the layer thickness, a vertical
variogram can be modeled for spatially varying properties immediately
after performing the log upscaling.
For the purpose of this exercise, it is the vertical range that is
important, so do not be too concerned about nugget, sill and other
parameters.
The main purpose is to extract an upper bound for the layer thickness
for the zone (i.e., ½ x vertical range).
The spherical variogram model will be used because the actual
range is equal to the effective range for that model.
The QC of the grid resolution should be done before you spend time
populating the property grids.

Property Modeling 3D Grid Quality Control Tools • 201

 The next slide illustrates how Petrel views the data when performing
most property modeling methods. This is called the SIMBOX mode. The
following three slides illustrate the basic procedure we will follow for
3D Grid resolution QC in the exercise using the provided data set.

3D Grid Quality Control Tools

Spatial Analysis – Simbox Mode
The experimental variogram can be visualized in
real or simbox mode.

Toggle simbox on and off by using Toggle

simbox mode icon.

For the horizontal direction:

- using upscaled logs and simbox mode will ensure
that only samples from equivalent geological layers
are compared.

For the vertical direction:

- for vertical resolution of the model, use the raw log
data with the simbox turned off.
- for modeling, use the upscaled logs with the
simbox turned on.

Simbox Mode
It is possible to visualize a property as a regular box in XYZ direction.
This option is also called Simbox view. When the property is visualized
as a regular box, all of the cells in the same layer get flattened and can
be easily compared, it means that the topography and the faults have
been flattened.
The use of upscaled logs and simbox mode is recommended for
horizontal analysis (Major and Minor direction variograms), as it will
ensure that only samples from equivalent geological layers are
compared.

202 • 3D Grid Quality Control Tools Property Modeling

For the vertical direction, it depends on what you plan to use the
analyzed results for. If you are using the Data analysis process to
decide the vertical resolution of you model, you may want to use the There should not be
raw log data with the simbox mode turned off, as you are interested in too much of a discrepancy
between the raw and
the “real” spatial variability from the log sampling to guide the right- upscaled data variogram
sizing the grid. If you analyze for Property modeling, it is ranges.
recommended to use the upscaled logs with the simbox mode on, as
Petrel will use the upscaled value together with simbox view when
executing the modeling for some algorithms.

3D Grid Quality Control Tools

Vertical Variogram Analysis – 3D Grid Resolution QC
Use raw log data as source
Normal Score transformation

Grid layering should be

<= ½ vertical range

Toggle off Simbox mode Lag spacing =

for vertical analysis 1-2 x the sampling

3D Grid Right Sizing

In the Data analysis process, the data has to be normalized (standard
normal distribution) and stationary (no spatial trend) before you
perform the Variogram analysis. In this example/exercise, a Normal
score transformation has to be performed to normalize the data.
Lag spacing is adjusted so that a lag of 1 is approximately 1-2 times
the natural sampling rate of the log curve data. When this is achieved,
Property Modeling 3D Grid Quality Control Tools • 203
 the variogram will facilitate the interpretation of a range in the vertical
direction that is based on the heterogeneity in the raw data (this is why
the raw data is used for this analysis). The goal is to ensure that a
proper 3D Grid resolution has been designed to enable this
heterogeneity to be captured in the property grid. If the grid layering is
too coarse (> 0.5 vertical range), you risk over-upscaling the log data
and losing the ability to capture this vertical variation present in a given
property.
The number of lags is adjusted by increasing the size of the search
window so that enough lags are computed so you can to study the
vertical behavior of the data across the entire zone (that is, the number
of lags = total distance / lag spacing).
You can look at the statistics for the zone (Zone filter of the 3D grid) to
find out the average vertical cell height present in a given zone
(Average Z inc).

3D Grid Quality Control Tools

Vertical Variogram Analysis – Right Sizing (correct layering)

No conditioning to facies; to find first

variogram plateau/peak

Grid layering should be <= 11 m

Search Radius & Number of lags

decide the Lag distance

Find a Lag distance approx. equal

to sampling distance (of a log)

204 • 3D Grid Quality Control Tools Property Modeling

Correct Right-Sizing
In the slide example, the vertical range can be interpreted as 22 m from
the variogram. Half of the vertical range in this example amounts to 11
m. This suggests that grid layering should not exceed 11 m in thickness.
The porosity data is used without any conditioning to facies to ensure
the samples provide the maximum number of data pairs at very short
lags (that is, starting at the data sampling rate). This guarantees that
we can see the correlation at the smallest observable scale with
regards to the log data. Without any conditioning you often see the
“hole effect”due to cyclic values (that is, stacking of upward fining or
coarsening sequences of petrophysical values). Under these conditions,
the first peak can be modeled with confidence that the minimum range
for the layer thickness selection has been located. If the same analysis
is done using the facies conditioning, it is harder to pick out the vertical
range.

3D Grid Quality Control Tools

Vertical Variogram Analysis – Right Sizing (incorrect layering)

Upscaled Cell Selected

Lack of Vertical Variogram Structure

Cell Thickness is too coarse!

Property Modeling 3D Grid Quality Control Tools • 205

 Incorrect Layering?
In the slide, the current layer resolution for the 3D Grid is 18.7 m
(Average zinc (along pillar). This is well beyond ½ the vertical range = 8
m).
Analyzing the data from the upscaled cells gives a lack of vertical
structure in our example. Notice that the vertical variogram generated
with the same lag parameters from the upscaled cell data cannot be
interpreted.
This experimental variogram fails to capture the same spatial
correlation as that of the raw data. The range is clearly seen in the raw
data, but totally unrecognizable in the corresponding upscaled data due
to too coarse layering.
The conclusion is that the current grid layer resolution is too coarse.
The Layering process must be redone with other values.
The vertical range for the upscaled data should always be >= that of the
raw data. This is because the upscaled data has been pre-averaged.
The whole idea of this exercise is to make sure that this upscaling has
not over-averaged the log data values.
How to check the layer resolution:
In the Layering process:
For the options Follow base and Follow top, the Cell thickness is
specified directly in the process dialog.
For the options Proportional and Fractions, only the number of cells
is specified, not the cell thickness. To find an average value for the cell
thickness for a given zone, right-click on the Zone in the Zone filter of
the 3D grid. In the Statistics tab, look for the Average zinc (along
pillars).

3D Grid Quality Control Tools – Exercises

QC of 3D Grid layer resolution
In the previous exercise, you examined the univariate statistics for both
the raw and upscaled petrophysical log data. Such comparisons
frequently reveal inadequate grid resolution when a histogram of
upscaled data is compared to that of the corresponding raw data. While
206 • 3D Grid Quality Control Tools Property Modeling
this semi-quantitative approach can uncover problems associated with
over-upscaling, it will not allow you to select the proper layer resolution
without some trial and error.
On the other hand, vertical variogram analysis computed using the raw
log curve data can be used to determine the range in the vertical
direction. The ½ of this distance can serve as the upper bound for the
layer resolution for a given property.
Clearly, different vertical ranges can be present for different properties.
In the end, a common definition of the grid layering will be used when
populating facies, porosity, permeability, and so on. For this reason,
preliminary vertical variogram screening should be conducted on all log
curve data you intend to use as input for the property modeling. This
will ensure that the final selected layer resolution can properly
accommodate the heterogeneity present in all of the log property data.
A casual selection of overly fine layering can cost precious project time.
Conversely, layer resolution that is too coarse may be a waste of time,
as the results may not adequately capture critical reservoir
heterogeneity.

This exercise has a pre-made Petrel project stored in Projects folder:

Property Modeling 2010_M7&[Link]. If you did the last exercise,
please use the same project.

Exercise Steps
1. Select the Exercise Model grid to activate it. Open the Data
analysis process and select the PHI (U) upscaled property.
Unlock the zone setting by clicking on the Leave settings
unchanged button for Zone A and the button will now

look like this: . Select raw log data .

2. In the Transformations tab, perform a Normal score
transformation as shown in the next figure (on, by default).

Property Modeling 3D Grid Quality Control Tools • 207

 3. Go to the Variograms tab and proceed immediately to the
Vertical direction tab.

4. First, select the Toggle simbox mode off .

5. The Search radius should span the entire zone. Adjust the No
lags (number of lags); the Lag distance will change
accordingly, so make sure a sufficient number of experimental
variogram samples are computed. Perform a preliminary
analysis for the porosity data values in Zone A as shown in the
the next figure.

208 • 3D Grid Quality Control Tools Property Modeling

Hint: Use a Spherical model type; for this type of variogram model,
the actual range and the effective range are equivalent. This makes it
easy to read the vertical range and compute the upper bound layer
This type of “right
thickness from this analysis. sizing” analysis is easiest to
perform in zones that use
6. Compare the vertical range for the raw porosity log data layering parallel to the top,
to the base or to some
to that of the upscaled porosity log data. Do you see any reference surface. For
significant differences in the vertical range for the raw and proportional layering, you
must use the vertical range
upscaled porosity data? to predict how many layers
are required to ensure the
layer thickness does not
exceed ½ this vertical range.

Property Modeling 3D Grid Quality Control Tools • 209

 7. Repeat a similar analysis for Zone B as shown in the following
The porosity values steps with associated figures.
in Zone B may be difficult to 8. Add a 1D trend transformation above to Normal score (it
model using only a Normal
score transformation. You
“removes” temporarily the vertical trend). Note that the zero
may want to apply a 1D values are not representative as they represent the shale
Trend transformation in the Z facies. Try to fit the modeled trend line (blue line) by
direction. This will not
remove any data, only split positioning the mouse pointer on the blue points and move
them into a regression line them according to the following image to model the trend.
(trend) and the residuals 9. Apply a Normal Score Transformation.
(without trend).

210 • 3D Grid Quality Control Tools Property Modeling

Property Modeling 3D Grid Quality Control Tools • 211
 10. Go to the Variograms tab>Vertical direction for the raw
logs in Zone B model the experimental variogram (see picture
below).

11. Now, change to upscaled logs to visualize the vertical

variogram in Zone B.

212 • 3D Grid Quality Control Tools Property Modeling

 12. Check the Vertical range for both zones and compare with
the Average zinc (along pillar) for every zone described in The histogram
the Statistics tab (Exercise Model>Zone filter>Main_ results for raw and upscaled
data may look similar, but
pay>Zone A/B> right-click>Settings>Statistics tab). the corresponding variogram
How is the grid resolution compared with the vertical ranges? analysis reveals differences
in vertical correlation.
Comments Despite the reasonable
If you discover that the current grid layering is too coarse, it is easy to histogram, this situation
suggests that the grid
redo the layering and upscale the log data again. Ideally, this QC step resolution may be affecting
should be completed prior to populating properties for either facies or the ability to retain the
petrophysical data. If, on the other hand, you find the current grid is spatial characteristics
exhibited by the raw data. It
finer than necessary, you do not have to worry about the layer thickness is desirable to retain both
and can proceed with few cells (just be aware that too many cells may the distribution and the
spatial characteristics before
increase CPU time, affecting display performance and disk space proceeding with Facies and
requirements significantly). Petrophysical property
Overall, spatial variability can be captured by using the Variogram modeling.
analysis, but if there are zones/layers of particular interest, like thin
lateral shales that act as barriers to flow, they must be captured in the
Zones/Layering part regardless of the suggested layer thickness from
the variogram analysis.
Property Modeling 3D Grid Quality Control Tools • 213
 Summary
Quality control tools, which involves checking statistics, upscaling and
also the 3D grid layer resolution from the structural modeling process of
Petrel, have been covered in this module. It has been demonstrated that
the vertical variogram is an excellent tool to check the layering in the
sense of spatial correlation based on raw log data.

214 • 3D Grid Quality Control Tools Property Modeling

Module 9 - Facies Modeling

This module gives an overview of Facies modeling in Petrel and

shows you how to build a realistic facies model, how to perform a
statistical discrete data analysis for preparing inputs and quality
control.

Prerequisites
To successfully complete this module, the user must have knowledge of
the following:
• Familiarity with Geostatistics fundamentals
• Familiarity with geology fundamentals: Basic Sedimentology
• Familiarity with Reservoir Modeling fundamentals
• Petrel Introduction

Learning Objectives
The purpose of this module is to give the participant the basics of
Facies modeling and become familiar with the related tools in Petrel. At
the completion of this training, you will be able to:
• Understand how to construct a 3D facies model
• Have an overview of Depositional environments which can be
used conceptually as 3D models
• Understand geostatistical parameters for performing proper
Facies modeling
• Use the Data analysis tools to prepare data and to perform
quality control
• Perform statistical analysis for discrete data (including how to
create variograms from data) to perform Facies modeling

Property Modeling Facies Modeling • 215

 Lesson

Facies Modeling
Facies Modeling is the population of discrete data, such as lithofacies,
into the cells of a 3D grid.
Normally, the user will have upscaled well logs with discrete properties
in the grid. It is also important to try to identify any trends in the data,
within the reservoir, by analyzing the input data in the Data analysis
process.
To build a realistic facies model, you must have an understanding of the
geological processes both during and after deposition. You must also
have some knowledge of the reservoir connectivity and the level of
facies heterogeneity. The input to the modeling has to honor the
different descriptive facies information, such as shape, size and
orientation of the different facies bodies.

216 • Facies Modeling Property Modeling

 Facies Modeling
Common Facies types – Carbonates

Facies Modeling
Depositional Environments - Carbonates

Carbonates are formed in shallow seas containing features such as reefs, lagoons
and shore-bars.
Carbonate porosity

Interparticle porosity

Intergranular porosity

Intercrystaline porosity

Moldic porosity

Property Modeling Facies Modeling • 217

 Carbonate depositional environments
A reef is the simplest type of carbonate deposition (like the
skeletons of reef animals). In shallow lagoons, Calcium Carbonate is
deposited with shells etc. In addition to the carbonates, these
environments also produce evaporites such as salt and anhydrite.
Changes in sea level allow the deposition of salt or anhydrite as a seal.
Carbonate porosity
• Interparticle porosity - Each grain is separated, giving a similar
pore space arrangement as sandstone.
• Intergranular porosity - Pore space is created inside the individual
grains which are interconnected.
• Intercrystalline porosity - Produced by spaces between carbonate
crystals.
• Moldic porosity - Pores created by the dissolution of shells, etc.
Vuggy porosity is also created by the dissolution of fragments, but
is unconnected.

Facies Modeling
Common Facies types – Clastics (simplified)

ONSHORE

SHORELINE/
TRANSITION

DELTA SLOPE
OFFSHORE SHELF

218 • Facies Modeling Property Modeling

Clastic depositional environments
Sedimentary rocks are formed from the material of other rocks which
could be igneous, metamorphic or older sedimentary rocks. Rocks which
form from materials transported are clastic rocks. Non-clastic rocks are
from chemical or biological origin with subsequent deposition.

Facies Modeling
Depositional Environments - Marine
The depositional environment can be lake or continental, shallow or deep water
marine. The environment determines many of the reservoir characteristics
Sea level

Distal
turbidite
Shelf
carbonate

Upper-fan
conglomerates Mid-fan
sands Basinal
Shales

The depositional environment is often a vital part of the evaluation of a

well and a field. This often defines the major lithology. Lithology can
vary, also over relatively short distances: The shallow fan of the delta in
the slide produces a conglomerate, while the deep water is showing
shales (fine sediments).

Property Modeling Facies Modeling • 219

 Facies Modeling
Depositional Environments - Continental
Continental deposits can be eolian sand dunes, alluvial fans etc. A shallow marine
environment has a lot of turbulence, hence varied grain sizes. It can also contain
carbonates and evaporites.
A deep marine environment produces fine sediments.

Sandstone porosity

The classical continental deposition of sand dunes produces an

excellent quality reservoir rock with deposition of sediments by wind in
eolian environments producing clean, well-sorted sandstones. To create
a reservoir, the dune has to be buried with a source rock and cap rock
providing the rest of the elements of the reservoir. Typical cap rocks
could be shales or evaporites (anhydrite). These may be laterally
extensive or patchy, thin or thick - and this is crucial to understand so
that the layering of the 3D grid is done correctly to capture the sealing
rock type.
Channels
Some types of deposition occur in rivers and sand bars.
The river forms a channel where sands are deposited in layers. Rivers
carry sediment down from the mountains which is then deposited in the
river bed and on the flood plains.
Changes in the environment can cause these sands to be overlain with
a shale, trapping the reservoir rock. Ancient river beds can add up to a
220 • Facies Modeling Property Modeling
considerable thickness, and although the individual river channels are
small, they are numerous leading to a commercial reservoir.
The shape of a river/channel type deposition is often complicated,
causing problems for the placement of wells. Dipmeter analysis can
help to determine the direction of paleoflow and hence can assist in
well placement. When modeling such channels, their direction and
interconnection is crucial for understanding flow in the reservoir.

Facies Modeling
What to look for when deciding on a Facies Model

Objective:
- Capture large scale heterogeneity
- Model facies architecture (flow units and barriers)

Modeling Technique:
- Deterministic or Stochastic
- If Stochastic; Pixel based
or Object based

How to build a facies model?

The goal is to build a 3D-model that captures the reservoir architecture
with flow units and barriers. Different flow units can be modeled as
zones while different barriers can be modeled as faults. Depending on
the environment of the deposits and the amount of input data, an object
or pixel-based method has to be chosen.
Three different stochastic pixel based methods are available in Petrel:
Sequential Indicator Simulation, Truncated Gaussian Simulation and
Truncated Gaussian with trends.
In general, with sparse data you should start with a Stochastic (pixel
based) model, based on the SIS or TGSIM methods. With more
available data and an idea of the conceptual model, an object based
Property Modeling Facies Modeling • 221
 model can be created. This is still stochastic but follows certain rules.
Objects that cannot be modeled can be drawn (deterministic) as a final
edit of the model.
Pixel based techniques: Sequential Indicator Simulation tends to
produce random facies (honoring upscaled data) but with indistinct
boundaries between the different facies.
Object based techniques: can produce sharp clean facies
boundaries. Such representations are often preferred due to the more
realistic appearance.
When upscaled for simulation, both approaches produce reasonable
results.
Practical exercises for facies modeling using both pixel and object
based techniques will be performed during this course.

Facies Modeling
Which Facies modeling approach should we use?
If well logs are up-scaled, they can be used in Deterministic and Stochastic methods
If no well logs are available, most Deterministic methods cannot be used (apart from
Assign values, Calculator and Interactive drawing), and mainly Unconditional
Stochastic methods are used.

Deterministic techniques
- Are typically used when dense data is available (many wells or wells + seismic)
- Yield a single estimated result

Stochastic techniques
- Are typically used when sparse data is present
- Can produce multiple equally probable realizations (outputs)

222 • Facies Modeling Property Modeling

Facies Modeling
Petrel Modeling Techniques for Discrete Properties

Main course objective: Stochastic methods

Pixel Based: Shape decided by Variograms, trends etc.
SISIM TGSIM TGSIM with trends

Object Based: Facies with defined geometric shapes

General object Fluvial Adaptive Channel

Facies modeling methods in Petrel

The current suite of facies modeling techniques that can be performed
in Petrel range from interactive facies modeling (paint brush approach)
to various stochastic methods capable of generating multiple
realizations.
The most widely used methods are Sequential Indicator Simulation
(SIS) and Stochastic Object Modeling. These methods are commonly
used and will be the main focus of this course. Also Truncated
Gaussian methods are commonly used and will be covered in the
course.

Property Modeling Facies Modeling • 223

 Facies Modeling
Exercise Data – Well Correlation and Facies interpretation

Off-axial Axial Off-axial

Well correlation - course data

By using the different log signatures, various facies types can be
identified. The assumption is that facies analysis has been performed or
will performed based on this knowledge.
The well correlation section display illustrates the course dataset,
showing facies interpretation of four different facies; shale, levee, lobe
and channel.
Both well DW5 and well DW6 are located in the off-axial part of the
reservoir and dominated by shale in both of the reservoir zones.
The remaining wells are located in axial regions and show coarsening
upwards sequence in the lower portion of Zone B corresponding to lobe
facies.
A deep water turbidite reservoir is:
• Generally progradational.
• The axial part is sandier and more channelized.
• The off-axial part is shalier and dominated by overbank facies.
224 • Facies Modeling Property Modeling
Facies Modeling
Exercise Data – Theoretical Model for Deep Water Turbidite Facies
Channel, Levee and Lobes

From Deep-Water Sandstones, Brushy Canyon Formation, West Texas, (Field Guide For AAPG Hedberg Field Research Conference - April 15-20, 1999)

Depositional environment - course data

A facies model can be based on a theoretical scheme derived from our
first well correlation attempt.
The course data set represents a deep-water turbidite reservoir. The
reservoir includes an axial part dominated by channel and levee facies.
The off-axial part is shalier and is dominated by over bank facies.
The slide shows a schematic paleogeographic map of a prograding
submarine fan system with typical deep-water turbidite deposits.

Property Modeling Facies Modeling • 225

 Facies Modeling
Exercise Data – Conceptual Sedimentological Model

Conceptual model - course data

This slide shows a schematic diagram of a conceptual model and the
goal is to capture this in digital form using the log curve data. The
course dataset is dominated by turbidite fan sands in the lower part
(zone B) and turbidite channel sands and levees in the upper part (zone
A).
Typically the Channels have a thickness-width relationship, and the
“gull-wing” Levees are normally only measured as a fraction of the
channel thickness.

226 • Facies Modeling Property Modeling

 Facies Modeling
Statistical Data Analysis

Data analysis is a process of data QC, understanding the data and preparing
inputs for Facies modeling:
Facies Proportion : vertical facies variation
Facies Thickness : thickness of individual facies intervals
Facies Probability : calibration with secondary attributes
Discrete Variogram : spatial facies continuity

Data Analysis - Facies modeling

Petrel provides statistical discrete data analysis functionality including
vertical facies proportion, facies thickness, facies probability and
discrete (indicator) variogram analysis. This functionality allows you to
quality check the input well data to understand statistical facies
variation, and to prepare input for facies modeling. The data analysis
results can, together with the conceptual sedimentological model, be
used in the facies modeling process to build a more realistic facies
model. This analysis is the first step in facies modeling and can be
performed once an upscaled facies property is available.

Property Modeling Facies Modeling • 227

 Facies Modeling
Facies Data Analysis – Vertical Proportion
Proportion : Applied as vertical probability curves based on the original fraction of
facies in each K-layer. The probability of using the given fractions is given by a curve
that can be edited manually.

Channel and levee facies

are abundant in the
Fit probability curve shallower portion of the
To fraction histogram interval, while Lobe
facies is concentrated in
the lower portion of the
zone

Original values (fixed) Probability values (editable)

Vertical Proportion
A method to investigate and edit the vertical distributions of each
facies in a selected zone. These values can be used automatically in the
Facies Modeling process to control the vertical distribution of the
different facies.
Sometimes it is sufficient to use the same fractions as given by the
Since you are using upscaled cells, but other times these are under or over-represented in
probability curves rather than the well position compared to the expected 3D population, and must be
real fractions, you may not
always get a 100% match of
edited. The edit can be performed per facies per K-layer.
what was set up. Also, other
factors like large variogram
ranges and trends will
interfere with the results.

228 • Facies Modeling Property Modeling

Facies Modeling
Facies Data Analysis – Thickness of Facies types
Thickness : Histogram view of thickness variations of each facies type. The bin
interval can be specified in project units to increase / decrease the resolution
Display upscaled / raw logs
Add/remove facies codes

With the given Bin

interval of 4 m:
Channel and levee facies
vary between 4 to 28 m.
Lobe facies varies from
12 to 32 m.

Thickness Histogram
This analysis allows you to review the thickness distribution for each
respective facies. This particular analysis is for informative purposes
only (no editing is supported). The information can serve as input to
subsequent facies modeling (for example thickness distributions for
object modeling). This analysis can also be used to check that the cell
thickness is at adequate resolution (for example checking that the layer
thickness are less than the minimum observed thickness from that of
the raw log curve data).

Property Modeling Facies Modeling • 229

 Facies Modeling
Facies Data Analysis – 3D Probability
Probability : Shows the relationship between the original upscaled facies and the
secondary attribute which must exist in the cells to be populated.
Secondary property (here:Acoustic impedance)
Facies type locked to
avoid editing

At low Acoustic impedance,

generating Channel, levee
and Lobe facies are almost
Edited point equally probable. As the
Anomalous point impedance values increase,
shale is more and more the
probable facies

Probability
A method to investigate and edit the relationship between a discrete
property (e.g. facies) and a continuous property (e.g. seismic attribute)
that has been sampled into the same 3D grid as the facies property.
This relationship can be used both with the Sequential Indicator
Simulation (SIS) and the Truncated Gaussian Simulation algorithms.
It is important that the secondary property has a good correlation
(positive or negative) to the facies. A common secondary property to
use is Acoustic impedance as it correlates well with porosity which
again correlates well with standard clastic facies types.
The background histogram show the amount of data and is good for
qc’ing statistics on anomalous points (these can then be edited). Ideally,
probabilities should be smoothed.

230 • Facies Modeling Property Modeling

 Facies Modeling
Facies Data Analysis – Indicator Variograms
Variogram: Variograms should be modeled for each facies as facies vary in
correlation length with distance. How bundled these stochastically modeled pixel
facies types appear is dependant on Variogram range and variance (nugget).

Facies type

Variogram results – to be
used in Modeling

Search cone setup – to generate

the sample variogram

Indicator Variogram
This analysis allows you to compute and model indicator (discrete)
variograms for each facies. Such variograms serve as input to the
Sequential Indicator Simulation (SIS) and to both the Truncated
Gaussian Simulation with and without trends algorithms. Anisotropy
(major direction and the ratio between the major and minor range) can
also indirectly provide valuable parameters for object modeling (actual
variogram is not used directly as an input in object modeling).
To generate a good variogram in all directions, it is critical to get an
understanding of the spatial variability of our data and how correlated
they are. When we use any variogram-related algorithm, we should
have a pretty good idea of the parameters of the Variogram ellipsoide.
Remember, that the variogram varies with different facies, lithologies
and rock types:

Property Modeling Facies Modeling • 231

 1. It does not matter whether you model the vertical or horizontal
variogram first. However, as the nugget and model type must be
the same for all three directions. It is better to start with the
vertical variogram since this is usually better defined because
the vertical well logs are densely sampled, and the nugget is easy
to find. The nugget represents the variance at a range below the
lag distance, or the smallest separation between points. Since this
can be considered to be anything down to zero separation distance,
the nugget has to be the same for all 3 directions.
2. If there is a lot of wells present and/or there are horizontal wells,
we may attempt to find horizontal variogram ranges. However,
due to lack of data this may be too coarse and even futile.
3. However, if any correlated data source has been found (like a
resampled acoustic impedance cube) and it exist in every cell
location to be populated, it can be used as a secondary variable
when doing stochastic modeling. Since it is used in the actual
modeling, we may also derive the remaining variogram parameters;
the major direction (if anisotropy), major and minor range. The
major direction can be found from a Variogram map, and the major
range can be found from the Sample variogram in the direction
indicated by the map. The minor range can be found 90 deg. to the
major direction. If the reservoir is homogenous and isotropic, the
major and minor ranges are equal.

232 • Facies Modeling Property Modeling

 Facies Modeling
Facies Data Analysis – Indicator Variogram calculation
Calculation process: Example:
1. Discrete data is transform as binary
variables. The facies of interest is converted
to ‘1’ and the remaining facies to ‘0’.
2. The classical variogram is calculated using
the binary codes to create the semivariance
for each lag for that discrete value :
Nh 2 Indicator variogram for Shale:
1
g (h ) =
2Nh
∑ (( facies ) − ( facies ))
(i + h ) i • Shale is recognized as ‘1’ and other facies as ‘0’ to
i =1 calculate the variogram (semivariance)
3. A prior distribution function (pdf= F(z)) is • The variogram is standardized by the pdf factor
calculated: i −1
F ( zi ) = ∑ P ( z j )
j =1

Where : P ( z j ) = facies proportion

4. The variance for a discrete property is
calculated according to the distribution
(Var= F(z)*(1-F(z))).
5. Then the classical variogram is
standardized by Var= F(z)*(1-F(z)):
(g (h ) )
[F ( z ) ∗ (1 − F ( z ))]
Indicator Variogram calculation
For the indicator variogram, the facies of interest will be converted to
1, the rest is 0. Then the variogram will be calculated with the classic
formula and at the end standardized by F(z)*(1-F(z)). In the slide, an
example of the calculation for the Shale facies is shown. Therefore,
sills above or below 1 can be obtained. The variance of a discrete
property is calculated according to the distribution. For example, the
Bernoulli distribution has the following variance: Var=p*(1-p) where p is
the discrete mean or, in Petrel, it is called fraction or proportion. The
similarity of the standardization of the discrete variogram can be
realized where the pdf (prior distribution function) is used instead.
In the example, pdf=F(z) is calculated:
F(shale) = Pshale = 0.50
F(sand) = Psand + Pshale = 0.05+0.50 = 0.55
F(silt) = Psilt + Psand + Pshale = 0.15+0.05+0.50 = 0.70
F(fine silt) = Pfine silt + Psilt + Psand + Pshale = 0.20+0.15+0.05+0.50 = 0.90
Property Modeling Facies Modeling • 233
 For the standardized variogram of the Shale facies:
Lag (i) = (Gamma (hi)) / (F(zi)*(1-F(zi))), where i=1,2...n
Lag 1 = 0.02/0.25=0.08
Lag 2 = 0.12/0.25=0.48
Lag 3 = 0.27/0.25=1.08
Lag 4 = 0.32/0.25=1.28
Lag 5 = 0.36/0.25=1.44
Lag 6 = 0.35/0.25=1.40

Extra: Advanced Property Modeling course

(2 days)

Advanced use of Pixel and Object

based methods
Hierarchical facies modeling
Theory on Multipoint Statistics (MPS)
Create Training images
Artificial neural networks
Seismic Geobody extraction and use
Facies Modeling
Multipoint Facies Simulation based on
Training images, Patterns, Soft/Hard data

Petrophysical modeling
Conditioning to Facies models

234 • Facies Modeling Property Modeling

Facies Data Analysis – Exercises
Vertical facies proportion analysis
Vertical facies proportion function allows you to visualize and edit
vertical facies variation. The original facies percentage is displayed
layer by layer to demonstrate the vertical facies distribution. In cases
where the vertical proportion calculated from available wells may not
be representative due to insufficient well observation, the curve can be
edited manually to ensure anomalies are not transferred to the model.
For your convenience, an extra project called Property Modeling
2010_M9.pet has been created with all necessary data in case you
have not completed any of the previous modules in the book. If you are
satisfied with your project, continue to work on it.
Exercise Steps
1. In the Models pane, make sure the Exercise Model is
activated.
2. In the Processes pane, double-click on the Data analysis
process.
3. Select Facies (U) as the property to analyze. Select Zone A
and deselect the lock button to open the function panel.
4. Select the Proportion tab. The left graph shows the original
facies percentage calculated by layer. The right graph is for
manual facies probability editing. In Zone A, there is no
obvious meaningful vertical trend observed and no need for
editing.
5. By default, the vertical proportion for each facies in the editing
graph is a constant value. To fit the curve to the variation
observed, click the Fit active / all curves to histogram
button.
6. Select Zone B and deselect the lock button. In the Proportion
tab there is a clear vertical trend showing that the channel and
levee facies are mostly in the upper part of the zone and the
lobe facies in the lower part of the zone.
7. Fit the curve again (click the Fit active / all curves to
histogram button).
8. Try to edit the proportion for Lobe facies (see figure) to make it
smoother, click the Smooth active / all curve button.
Property Modeling Facies Modeling • 235
 9. Click on Apply, the result can be used as input for later facies
modeling.

Facies thickness analysis

The facies thickness function allows you to visualize the facies
thickness distribution as a histogram (see next figure). You can select
different facies and different bin intervals. The display and statistics
are read-only. The statistical numbers calculated should be used with
caution if they are being used as input for facies modeling as the
thickness is calculated from well facies intervals. These intervals can
be stacked facies bodies. Also, the facies thickness input in facies
modeling is the center thickness of the body. A well can drill through a
facies body at any position. So the individual facies thickness input for
facies modeling generally should be higher than the well observation.
Exercise Steps
1. Continue from the previous exercise. Go to the Zone A, select
the Thickness tab in the Data analysis dialog.
236 • Facies Modeling Property Modeling
 2. Change the Bin interval to get a natural looking histogram.
Select the facies you want to look at. View the histogram and
statistics and compare the thickness distribution for the
different facies. Notice that the channel facies is generally
thicker than the levee facies (as discussed in the Lesson part).
3. Change to Zone B and see the thickness distribution for
different facies. Since there are too few samples for channel
and levee facies, the statistics calculated here need to be used
with caution.

Property Modeling Facies Modeling • 237

 Facies probability analysis
Facies probability analysis allows you to calibrate any continuous
property to facies observed from wells. For example, a seismic attribute
cube can be converted to a facies probability cube for each facies under
investigation. To use this function, the seismic must be re-sampled into
the same 3D grid as the facies property. The result can be used in a
Sequential Indicator Simulation.
Exercise Steps
1. Continue from the previous exercise. Select Zone A and go to
the Probability tab in the Data analysis dialog.
2. Select the Secondary attribute to be used (AI) to calculate
facies probability. The graph will show the facies probability of
each facies at a certain attribute value interval.
3. By default, the facies probability for each facies is a constant
value. To see the variation of facies probability with seismic
attribute, select a facies type and click the Fit active / all
curve(s) to histogram button.
4. There may be irregular points on the facies probability curves
due to data availability and quality (check histograms). The
facies probability curves can be edited / smoothed to make it
more natural and representative.

238 • Facies Modeling Property Modeling

 5. Select Zone B and go through the same process.
6. Click on Apply, the results are ready to be used as input to
later facies modeling.

Variogram analysis
Variogram analysis is used to determine the spatial variation of the
different facies. The process includes experimental variogram
generation and variogram modeling. It is an interpretative process and
should take into account geological knowledge. Remember that a
Variogram model normally needs to be defined for each facies.
Usually, vertical variogram models have plenty of data and can easily
be estimated. Horizontal variograms however, normally have very little
data, which may result in questionable sample variograms and are
consequently unreliable Variogram models. In such cases you can use
additional geological information, such as expected continuity of the
facies or a preferred direction (anisotropy), to define the ranges,
azimuth angles and nuggets for each facies.
Exercise Steps
1. Continue from the previous exercise. Go to the Zone A,
Variograms tab in the Data analysis process dialog.
2. Select the facies Shale.
3. Select a variogram direction: Major direction, Minor direction
or Vertical direction. Remember that the variogram analysis
for modeling must be done using upscaled cells and simbox
mode. The search cone area has to be big enough for data
searching. Starting with the variogram in the Vertical
direction make easy to define the nugget.
4. Define the Major direction to 134 degrees and open the
searching parameters as shown in the figure below. The value
for the major direction has been determined by the previous
anysotropy analysis (Module 2 - Basic Statistics Part 2
exercises).

Property Modeling Facies Modeling • 239

 5. The number of wells is probably not sufficient for calculating a
reliable sample Variogram for each facies. Determine the range
for the Major direction by fitting the variogram curve to the
sample variogram. The variogram ranges will be automatically
updated.
6. Model the Variogram for the Vertical direction as shown on
the next figure:

240 • Facies Modeling Property Modeling

 7. Use the same Variogram model parameters for all facies by
selecting the Same variogram for all facies check box or
by copying the settings from one facies and pasting them to

the other facies using .

8. Use the Copy and Paste tools from the Zones, to copy the
variogram parameters in the Zone B as well. Click on Apply/
OK to save the Facies data analysis.
9. For practical purposes, for the following exercises, all facies
can be given the same variogram according to the table below:

Facies Major range Minor range Vertical range Azimuth

Shale, Channel, Levee 2500 800 14 134

Property Modeling Facies Modeling • 241

 Summary
Several Facies modeling approaches were covered in this module.
Indicator variogram calculation and how to perform a statistical discrete
data analysis for preparing data for 3D modeling was also covered.

242 • Facies Modeling Property Modeling

Module 10 - Sequential
Indicator Simulation

Introduction
The basic concepts and use of Sequential Indicator Simulation (SIS) as
a pixel-based facies modeling algorithm to produce multiple realizations
will be covered in this module.

Prerequisites
To successfully complete this module, the user must have knowledge of
the following:
• Familiarity with Geostatistics Fundamentals
• Familiarity with Geology Fundamentals: Basic Sedimentology
• Familiarity with Reservoir Modeling Fundamentals
• Petrel Introduction

Learning Objectives
The purpose of this topic is for the user to gain basic knowledge of
Facies modeling using the Sequential Indicator simulation algorithm in
Petrel. At the completion of this training, you will be able to:
• Understand the principle of Sequential Indicator Simulation
(SIS) algorithm.
• Run the SIS method to generate facies models with multiple
realizations.
• Run the SIS method using results from the Data analysis
process.
• Learn how to make Trend data to guide the facies models
generated from SIS.

Property Modeling Sequential Indicator Simulation • 243

 Lesson

Sequential Indicator Simulation

Petrel Modeling Techniques for Discrete Properties

Stochastic methods used in this course:

Pixel Based: Shape decided by Variograms, trends etc.
SISIM TGSIM TGSIM with trends

Object Based: Facies with defined geometric shapes

Object Fluvial AdaptiveChannel

Facies modeling methods in Petrel

A widely used method is Sequential Indicator Simulation (SIS),
which will be covered in this course. As it is the first method many
modelers use when starting a project, it will be covered first.
After more data is gathered the modeler can move to other methods
like object modeling, but SIS still has a wide application also in more
constrained environments.

244 • Sequential Indicator Simulation Property Modeling

 Sequential Indicator Simulation (SIS)
Overview

SIS is a stochastic (pixel based) modeling algorithm, using upscaled cells as basis
for fraction of facies types to be modeled. The variogram constrains the
distribution and connectedness of each facies type. It is widely used to model
facies with unclear or undefined Shapes, or when few input data are available.

Inputs:
– Facies proportion, facies probability and 1D, 2D, 3D Trends
– Different variogram for different facies
Underlying methods:
– Simple Kriging (Global mean - stable)
– Ordinary Kriging (Re-estimates mean – more data)
Output:
– Pixel based property honoring input data
– Stochastic; multiple realizations can be run for Uncertainty

Sequential indicator simulation (SIS)

SIS is a stochastic modeling technique where typical inputs are:
• Upscaled well log data
• Defined variogram
• Random seed
• Frequency distribution of upscaled data points
• Trends in 1D, 2D or 3D
Petrel uses the GSLIB algorithm to perform this simulation. The
method is stochastic, so if the seed number is left to vary, the
subsequent realizations will be different. It is a pixel-based facies-
modeling algorithm used to model facies without clear shape and
boundary. In the early stage of the project, facies architecture, shapes,
and dimensions may not be clearly understood. SIS can be used to
generate a preliminary facies model and allows a stochastic distribution
of the property, using the input data distribution (seen in a histogram).
Directional settings such as variogram and extensional trends are also
honored.  
Trends and attributes can be used to help distribute facies between
well locations.

Property Modeling Sequential Indicator Simulation • 245

 Sequential Indicator Simulation (SIS)
When can we use SIS?

SIS is typically used for various depositional environments, especially with few
available input data (wells). Other considerations are:
Seismic
– Once a 3D seismic survey is available with correlated
attributes, SIS is strong as it easily integrates:
• 3D probability trends from seismic
• Attribute Probability from seismic in Data analysis
• Horizontal variogram ranges derived from resampled
seismic
Facies environments
– Useful in Carbonates which do not have defined
shapes or direct facies relationships
– Clastic environments where connectivity and geometry
of facies objects are not/yet determined

Sequential Indicator Simulation (SIS)

Theory

Cell (X3) is chosen along a random path X0

(decided by Seed number).
X3
A PDF (probability distribution function) is X1
derived from IK (Indicator Kriging)
Upscaled and simulated cells are used for
X2
facies probability calculation.
A simulated value (Shale) is drawn from the
Original probability of Sand/Shale given
PDF curve using random MC sampling (0-1) by Upscaled well logs:

Local indicator Psand= 0.3 & Pshale= 0.7

p
PDF curve
1.0 Cell at location (X0) to be simulated

0.45 Upscaled cell (Shale)

0.3 Upscaled cell (Sand)

Simulated cell (Shale)
Original probability at (X0) Simulated cell (Sand)

246 • Sequential Indicator Simulation Property Modeling

Sequential Indicator Simulation - Theory
Goal: The idea is to use scattered well lithology data (here, sand and
shale) to predict the probability of lithologies away from the wells.
Procedure: Using the SIS method, each grid cell is visited sequentially,
following a random path (given by the seed number). For each new cell,
the local sand/shale probability distribution (PDF) is calculated using
upscaled data and already simulated cells as control points. An
estimate of the sand/shale conditional probability is estimated using
Indicator Kriging (IK). The estimate is given by the variogram and
subsequently the variogram weighted linear combination of data from
the upscaled cells. A simulated value (sand or shale = 0 or 1) is drawn
at random from the local PDF curve.
Finally, the simulated cell value will be used for the next cell to update
the PDF as an additional control point together with the “hard data”
(upscaled cells).
Example from the previous slide:
Given sand – shale facies by upscaling the well logs. Calculate the
facies at location X0.
Indicator kriging (IK) calculates the probability of sand psand and of shale
pshale at X0. From the probabilities a pdf is calculated:
Draw a random number between 0…1 and determine the facies from
the pdf:
For example a Random number: 0.45 --> Shale
For SIS, a small
variogram range increases
the influence of the Global
Probability Curve.

Property Modeling Sequential Indicator Simulation • 247

 Sequential Indicator Simulation (SIS)
Setup in Petrel
1. Property and zone selection
A. Make sure to pick the correct property;
must be upscaled, i.e. have (U) as
suffix
B. Select SIS as method for one zone

2. Facies:
A. Select the facies from the template
B. Use the blue arrow to insert

3. Variogram:
A. Specify Range, Nugget and Type
B. or get a variogram from Data Analysis

4. Fraction:
A. Use Global fraction of Upscaled cells
B. or use probabilities (property/trend)

Sequential Indicator Simulation - Setup in Petrel

When using the Sequential Indicator Simulation algorithm as a default,
the fractions from the upscaled well logs are used. However, these can
be changed in the Fraction tab.
The model type to use for the variogram (exponential, spherical and
Gaussian) together with the range, nugget, and azimuth can be defined
under the Variogram tab.
When using results from Data analysis, the corresponding Data tab in
this dialog will be grayed out. There are three buttons that are linked to
the Data analysis results:

Variogram button to use the variograms made in the Data

analysis process.

Attribute Probability curves button to use the Attribute

Probability curves made in the Data analysis process.

248 • Sequential Indicator Simulation Property Modeling

Vertical proportion curves button to use the Vertical Proportion
curves made in the Data analysis process.

Sequential Indicator Simulation (SIS)

Result

SIS is a kriging-based stochastic method:

• Honor the facies distribution.
Small Range Large Range
• Honor the upscaled cells. Omni-directional Directional
• Distribute the facies in a “fuzzy manner”.
• There is no facies relationship.
• The amount of ‘bundling’ of a facies depends
mainly on the Variogram and trend input data.
• Multiple realization can be run for Uncertainty

Property Modeling Sequential Indicator Simulation • 249

 Sequential Indicator Simulation (SIS)
Global Facies Distribution Control

From Data analysis:

• Attribute Probability curves
• Vertical Proportion curves

From Well data or manually entered:

• Based initially on upscaled facies
• Facies logs directly or manual input

From Probabilities:
• Probability property cube (3D trend)
• Probability surface (2D trend)
• Vertical probability function (1D trend)

Global fraction
To start with, it is common to use the upscaled cells from well logs as
the Global fraction. When more data is available, the Global facies
distribution can be changed. For example, it can be done manually if
the proportion of sand is too high from the upscaled logs (for example,
are sands normally targeted when drilling).
However, instead of using the Global fraction from upscaled logs,
probabilities can be set from a cube (another property) or trends in
either vertical direction (function) and/or horizontal direction (areal
probability maps).
If the different alternatives for defining the fraction mentioned earlier
are used, the Trust fraction or trends option can be applied. By
Read the Tips button default, the total fraction and the maximum probability will be scaled to
and Online Help for 100%. Selecting this option will disable it, although the scaling must be
detailed information about
the Trust fraction or
done if these values go over 100%. The same applies for the trends.
trends. Trends are normally in range between 0 (0%) to 1 (100%).

250 • Sequential Indicator Simulation Property Modeling

If a proper Data analysis has been performed, the button options for
using either the Attribute probability curves from a correlated re-
sampled seismic property can be used. Alternatively, the Vertical
proportion curves can be used together with a 2D probability surface.

Sequential Indicator Simulation (SIS)

How to generate 1D and 2D Trend input

1D Vertical Trend 2D Horizontal Trend

Initial polygons
Define function

Z-value selector

Truncation
Sand Probability
Probability (0-1)

Y
Final trend map

X
Depth of zone used by the vertical trend function

How to generate Trend input in Petrel

1D trend
Generate a vertical probability function:
1. Make an empty folder and give it a relevant name (for
example Functions).
2. Right-click on the folder and select Create new function
from the pull-down menu.
3. Define Name, X and Y (min/max), and number of points for
the new function in the pop-up dialog. The initial function will
be linear, but it can be edited once created. Note that the
X-axis refers to the top and base depth in the part of the 3D
grid where the function should be applied.
Property Modeling Sequential Indicator Simulation • 251
 4. Click OK , The Settings dialog for the new function will be
displayed.
The Settings dialog of the function comprises four tabs: Style, Info,
Statistics and Function. The tab of interest is the Function tab in
which the curve can be edited:

Select this tool to edit/create new points along the curve (in
X,Y).
Select this tool to move the whole line (in Y).
The X and Y axis can be swapped by clicking the button.
The curve can be set as a log X and/or log Y function.
The curve can be viewed with or without using spline curves.

2D trend
Generate a horizontal probability map:
1. Digitize lines representing the contours of the trend map using
the Make/edit polygons process.
2. Enter values (between 0-1) to the contour lines using the
Z-value selector tool
3. Grid the lines in the Make/edit surface process to produce a
2D grid.
4. Under Post proc tab, set Min Z-value to be Truncated = 0
(not %).
5. Click Apply and QC the new Trend map which is stored in the
Input pane.
This trend/probability map can later be used as input to the Facies or
Petrophysical modeling processes.
Surface operations and the surface calculator, as well as the interactive
surface editing options, can be used to modify the trend surface.

252 • Sequential Indicator Simulation Property Modeling

Sequential Indicator Simulation – Exercises
Influence of Variogram parameters on a Facies model
By using Sequential Indicator Simulation, we will calculate different
facies models to analyze the influence of the Variogram model
parameters.
For your convenience, an extra project called Property Modeling
2010_M10.pet has been created with all necessary data in case you
have not completed any of the previous modules in the book. If you are
satisfied with your project, continue to work on it.
Exercise Steps
1. In the Models pane, make sure the Exercise Model is
activated and open the Properties folder.

2. Open the Facies modeling process under Property
modeling in the Processes pane.
3. Select the Facies_Variogram property (it is a copy of the
Facies(U) property). Select Zone A and deselect the Lock
button to open the function panel for this zone ( ).
Select the Shale, Channel and Levee facies by using the
Add facies button .
4. Keep the default variogram parameters as shown in the next
figure. Click Apply to run the modeling process.

Property Modeling Sequential Indicator Simulation • 253

 5. Display the new Facies_Variogram property in the 3D or 2D
window. Use the filter icons ‘I’, ‘J’, and ‘K’ on the right side of
the Petrel canvas to scroll trough the model using the Property

player .
6. Just as a test of your variogram knowledge: Change the
Variogram parameter according to the following list. Compare
the results and note the influence of the different parameters.

Major Dir. Minor Dir. Nugget Vertical Azimuth

10000 10000 0.1 10 0
10000 1000 0.1 10 0
10000 1000 0.9 10 0
10000 1000 0.1 30 0
10000 1000 0.1 10 -45

Comment
Make sure that all facies get the same Variogram parameters. Set the
indicated variogram parameters, the option Same variogram for all
facies is selected by default or select each facies and use the copy
and paste buttons to copy the Variogram parameters across,
but make sure that every facies gets the correct fraction under the
Fraction tab.
However, in nature most facies have different variogram parameters.

Influence of ‘Global facies proportion’ on Facies modeling

In the following exercise, the influence on facies modeling of both the
Global facies fraction and the Layer facies fraction will be shown.
Exercise Steps
1. Continue in the Facies modeling process. Activate the
property Facies_Variogram for Zone A.
2. Keep the default Variogram parameters as shown in the
previous figure or Reset to default settings by using the icon
.
3. In the Facies tab, go to the Fraction sub-tab and change the
Global fraction for Shale to Manual 20 % according to the
next figure.

254 • Sequential Indicator Simulation Property Modeling

 4. Select the Trust fraction or trends check box and click
Apply. Using this option all the fractions will be rescaled
according to the new Shale 20% fraction.

The remaining 80% fraction will be divided proportionally to the original

fraction of Channel (13.33 %) and Levee (16.19 %):
Channel original fraction + Levee original fraction = 29.52 %
Channel rescaled fraction = (13.33*80)/29.52=36.12% approximately
Levee rescaled fraction = (16.19*80)/29.52=43.88% approximately
Then Shale (20%) + Channel (36.12%) + Levee (43.88%) = 100%
5. Display the Facies_Variogram model and check the These proportions
may vary slightly depending
proportion of the different facies in the Settings>Disc. stat. on your variogram settings
tab (discrete statistics) and Histogram tab. Also, check the (closer if ranges are small).
Petrel message log.

Property Modeling Sequential Indicator Simulation • 255

 6. Set the Global fraction of the facies back to the Upscaled
cells.
7. Open Data analysis from the button on the Function
bar or from the Facies modeling dialog.
8. Make sure that Facies_Variogram property and the Zone A
are selected. Set the proportion of Shale to zero for the top
two layers by editing the probability curve as shown in the
next figure. Click OK.

256 • Sequential Indicator Simulation Property Modeling

 9. In the Facies modeling dialog, click the Use the vertical
proportion curves made in the Data analysis button.
10. In the Variogram tab, use the Spherical as Variogram type.
Leave the range of 1000 m for the Major dir and the Minor
dir. Enter a Vertical range of 5 m. Click Apply.
11. Using the K-filter , scroll through the Facies model and
check the Shale proportion for the top 2 layers of Zone A.
Notice that its proportion is almost fulfilled but nevertheless some
remaining parts contain Shale. This is due to the fact that it is only a
probability that is used in simulation (not the direct fraction), in addition
to the influence of Variograms and Trends (here the variogram ranges
can be reduced, then the proportions are honored more correctly).
12. Open Data analysis using the button and select the
Facies_Variogram property and ZoneA and click the Fit
active / All curves to histogram button to reset the
facies distribution.
13. Click OK.
Property Modeling Sequential Indicator Simulation • 257
 Using Variograms and Trends for Facies model generation
The following exercise consists of two parts. First, we will calculate a
facies model based on different variogram models for each facies.
Afterwards, we will simulate the facies guided by a 2D trend using the
same variogram model. Finally, we will compare the two results.
Exercise Steps
1. Continue using the property Facies_Variogram.
2. Deselect the Use the vertical proportion curves made in
the Data analysis button.
3. Use the Variogram model parameters you defined previously
in Data analysis by clicking the Use the variograms made

in the Data analysis button. Alternatively, you can use

the values from the table below (for practical reasons the same
variogram will be applied to all facies):

Facies Major range Minor range Vertical range Azimuth

Shale, Channel, Levee 2500 800 14 134

4. Check the facies fraction under the Fraction sub-tab and keep
the default values calculated from the Upscaled cells.
5. Click on Apply to run the model and check the running
Vertical Variogram
ranges can take facies message for the facies fraction.
thickness as reference. 6. Check the Disc. stat. tab in the Facies_Variogram property
Lateral variogram ranges are
normally determined by
Settings.
existing geological 7. Visually check the model through the I, J, and K filter.
knowledge in cases of data
insufficiency.
As you can notice, the facies are oriented in the desired direction.
However, all of the facies are evenly distributed throughout the model,
which is not realistic in this case.
8. Copy-paste the property Facies_Variogram and rename it
to Facies_trend.
9. Open the Facies modeling process for Facies_trend.
Continue using Use the Variograms made in the Data

analysis button (or, use the parameters given in the table

under point 3 of this exercise).
10. For Channel facies, go to the Fraction sub-tab and use the
Horizontal trend and Property trend options as Global
258 • Sequential Indicator Simulation Property Modeling
 fraction.
11. The trend surface is located in the Input pane, into the folder
named Property Modeling Input. Select the Channel_ind_
trend surface and use the blue button to drop it into the
process.
12. For the Levee facies use the Horizontal trend and Property
trend as well. Insert the Levee_ind_trend surface from the
same folder.
13. Select the Trust fraction or trends check box at the top of the
panel for the Channel and Levee facies.

When a trend is used, the initial global fraction will be ineffective. In

this case the probability is calculated as:
Probability (facies i)=(Global fraction i x Horizontal trend at X,Y)
/ (Mean horizontal trend)
If both Channel and Levee trusted fractions (trends) are:
• <1, only the non-trusted ones (Shale) will be rescaled to make
sum=1
• >1, these fractions (Channel and Levee) will be rescaled
proportionally to make sum=1 and the non-trusted ones (Shale)
will be 0

Property Modeling Sequential Indicator Simulation • 259

 14. Click on Apply to run the model. Check the result through I, J,
K filters and compare to the model generated without using
trends.
Optional: Crosscheck the result by making an average facies map.
1. Copy and paste the property Facies_trend, rename it to
Facies_trend_check.
2. Use the Calculator for ‘Properties’ and change all other
facies codes to 0 except Channel by typing:
Facies_trend_check = If (Facies_trend=1, 1, 0).
Click ENTER.
3. Open the Setting for the new Facies_trend_check property.
Go to the Operations tab and select Make average map as
shown in the next figure. Click Run to generate the map.
4. Visualize the created surface map stored in the Input pane,
and compare it to the input trend map Channel_ind_trend in
a 2D window.
Now, the facies distribution has the desired orientation and trend.
However, the facies shape and architecture are not captured.

260 • Sequential Indicator Simulation Property Modeling

Sequential Indicator Simulation with seismic attribute
Seismic attributes can be used to constrain Facies modeling when using
the Sequential Indicator Simulation algorithm in Petrel. To use a seismic
attribute cube for Facies modeling, this seismic cube has to be re-
sampled into the current 3D grid. Calibration is also needed between
the upscaled facies log and the seismic attribute during the Data
analysis process using the facies Probability function.
Exercise Steps
1. Copy and paste the Facies_variogram property and rename
it to Facies_ind_seismic.
2. Open the Facies modeling process dialog and select Facies_
ind_seismic as the property.
3. Check the Data analysis from the button to make sure
the facies probability analysis has been performed in the
Probability tab and click OK.
4. In the Facies modeling dialog, click the Use the Attribute

Probability Curves made in Data analysis button.

5. Use the Same variogram for all facies and input variograms
parameters according to the table below:

Facies Major range Minor range Vertical range Azimuth

Shale, Channel, Levee 2500 800 14 134 (-46)

Property Modeling Sequential Indicator Simulation • 261

 6. Click Apply to run the model and check the running Petrel
message log for the facies fraction.
7. Cross-check the result against the input seismic attribute AI in
Data analysis>Probability tab to see the relationship.
Display the AI property on a separate 3D window and Tile
vertical the two 3D windows and use the Property IJK
filter player.

262 • Sequential Indicator Simulation Property Modeling

Summary
In this module, Sequential Indicator Simulation, a pixel based stochastic
algorithm which can be used with or without secondary data, was
applied to populate facies in a 3D grid.
We also covered how to perform a quality check of the results, and we
looked at how secondary data (like resampled seismic attributes) can
be helpful through the Data analysis process results while performing
Facies modeling using Sequential Indicator Simulation method.

Property Modeling Sequential Indicator Simulation • 263

Module 11 - Object Facies
Modeling

In this module, the basic concepts of Object Facies modeling in Petrel

will be covered, such as adaptive channel modeling, general object
modeling and hierarchical facies modeling.

Prerequisites
To successfully complete this module, the user must have knowledge of
the following:
• Familiarity with Geostatistics fundamentals
• Familiarity with Geology fundamentals: Basic sedimentology
• Familiarity with Reservoir modeling fundamentals
• Petrel Introduction

Learning Objectives
In this module, you will learn:
• How stochastic object modeling works in general.
• The different types of objects available in Petrel.
• How to do hierarchical facies modeling by use of multiple
object models.

Property Modeling Object Facies Modeling • 265

 Lesson

Object Modeling
Petrel Modeling Techniques for Discrete Properties

Stochastic methods
Pixel Based: Shape decided by Variograms, trends etc.
SISIM TGSIM TGSIM with trends

Object Based: Facies with defined geometric shapes

General object Fluvial Adaptive Channel

Facies modeling methods available in Petrel

Stochastic Object modeling is a commonly used method when more
data is available or if strictly conceptual models/trends should be made.
For object modeling, there are three main types of objects to model.
Two metods are related to channelized environments; the fluvial and
adaptive channel bodies. Simple-shape objects can be modeled using
the general object body.

266 • Object Facies Modeling  Property Modeling

 Object Facies Modeling
Overview
Objective:
Realistically capture facies architecture and geometry
Methods:
Channel and isolated objects fully integrated
Adaptive channel modeling honoring wells
Hierarchical modeling within existing facies
Input:
Well data
Modeling geometry and shape rules
Vertical and lateral trends
Output:
Defined facies object models
Multiple realizations

Object modeling overview

A detailed integrated study can provide a better understanding of the
facies model. Object-based facies modeling can then be used to capture
more geological details such as shape, geometry, orientation etc. to
build a more realistic model.  
Object modeling allows users to populate a discrete facies model with
different bodies of various geometry, facies code and fraction.
Where the objects are not placed, a background facies can be assigned.
This can be a given facies code or an existing facies model. Different
erosion/replacement rules can be applied to different bodies. Vertical
and aerial trends can be used as options for defining the spatial
distribution.

Property Modeling Object Facies Modeling • 267

 Object Facies Modeling
Adaptive Channel Modeling
Overview of method:
Adaptive channel modeling is object-based and facilitates full well control through
use of the GRFS algorithm
History:
Introduced in Petrel 2007.1 for channels
Modified in Petrel 2008.1 to honor channel-levee
In Petrel 2009.1 a new feature has been added to
channel modeling conditioning to a Volume
In Petrel 2010.1 the method is based on GRFS
algorithm
Advantages:
Superior to traditional object modeling techniques in
situations with large numbers of well constraints
Provides controlable channel connectivity
Stochastic, providing multiple realizations

Adaptive channel modeling

Adaptive channel modeling is an object based facies modeling
technique which models channels with substantial well control.
Traditional object modeling techniques model the channel first and then
try to place it in the model. If the wells do not fit, the channel is
rejected. The adaptive channel algorithm uses Gaussian random
function simulation, taking all of the well data into account when
creating each channel. This means that channels are never rejected.
The adaptive channel modeling algorithm aims to better cope with large
numbers of well constraints. The speed of creating a facies model
depends only on the number of channels created. Conventional object
modeling techniques will be faster in situations with few well
constraints, and slower, and possibly even fail, with a lot of well
constraints.
In summary, Adaptive channel modeling should be used because:
• It is fit for almost any type of channel modeling.
• It is superior when large number of hard data must be honored.
• It provides channel connectivity through well data.
268 • Object Facies Modeling  Property Modeling
 Object Facies Modeling
Adaptive Channel Principle

Identify the wells through which the

channel must pass (A and B).
At the A and B wells, assign elevation
A C
values that are close to the elevation of the
plane to honor the well data.
B
At the wells that must be avoided (C),
assign a value that differs greatly from the
elevation plane.
Generate the gaussian surface which will
honor the predetermined point.
Intersect the surface with the inclined plane
and find the channel trajectory.

Modeling principle
The method to model a single adaptive channel is based on the
principle of generating an irregular surface and intersecting it with a
dipping plane. If the irregular surface is appropriately produced, its
intersection with the dipping plane is a curve which approximates the
shape of a channel. Of course, you will never see or need to explicitly
model the irregular surface or inclined plane. You only control the
channel geometry and data honoring. The surface and plane modeling
takes place conceptually “behind the scenes”.
In order to model the irregular surface, the algorithm performs a
Gaussian random function simulation (Petrel 2009.2 and previous
versions use the SGS algorithm). This method generates a surface that
honors predetermined elevations at certain points. These
predetermined elevations are assigned at the wells to guarantee that
the channel will pass through the wells that need to be connected,
while avoiding those that do not contain the given channel.
Basic principle of the algorithm:
1. Identify the wells through which the channel must pass (A and
B in the slide) and those that the channel must avoid (C in the
slide).
2. At the wells through which the channel must pass (A and B),
Property Modeling Object Facies Modeling • 269
 assign elevation values that are close to the elevation of the
plane to honor the well data.
3. At the wells that must be avoided (C), assign a value that
differs greatly from the plane elevation.
4. Generate the Gaussian surface which will honor the
predetermined points.
5. Intersect the surface with the inclined plane and find the
channel trajectory.
The details on how the geometry of the plane and the Gaussian surface
are determined in order to produce a channel with the desired
amplitude, sinuosity, wavelength, etc. are beyond the scope of this
course.

Object Facies Modeling

When can we use Adaptive Channel Modeling?
Adaptive Channel method is typically used for Channelized environments with a
lot of well control, but can be used for modeling any channel types:
Wells
– A large number of wells will yield a better
Braided river
control of:
• Time lines in correlation (sequence
stratigraphy)
• Channel connectivity through Tracer or
Well interference test analysis
Facies environments
– Clastic environments where connectivity
and geometry of facies objects are known
– Could be fluvial (onshore) or turbidites
(offshore)

270 • Object Facies Modeling  Property Modeling

 Object Facies Modeling
Adaptive Channel Modeling – Setup in Petrel

1. Property and zone selection

A. Make sure to pick the correct property; must be
upscaled, that is, have (U) as suffix.
B. Select Object modeling as method for a Zone.

2. Object type:
A. Add a new Adaptive channel using button.
B. Settings will appear in tabs after selection.

3. Facies:
A. In the Settings sub-tab, specify the correct Facies
for Channel and Levee.
B. Use the upscaled Fraction (%) or edit it.

4. Connectivity:
A. Use a Body index property to indicate wells that
should/should not connect.
B. A Layer tolerance can be defined for cross-
layer channels.

Adaptive channel modeling - Setup in Petrel

The algorithm will first honor the hard data in the wells, then the body
property, then geometrical information, and lastly global fraction
information. If it is not possible to honor the body property due to
geometrical constraints, the geometrical information will be overridden
and you will receive a warning in the message log. The message log
will tell you how each channel is made, the starting point for the
channel in the IJK cell position, the body number, the direction of the
channel, its amplitude, its thickness at the thickest point in cells, the
number of control points (cells where we know the channel either is or
is not), and lastly, the number of channel observations in the body.
Adaptive channels can be combined with fluvial channels and geometric
facies bodies, as well as with pixel-modeling algorithms.

Property Modeling Object Facies Modeling • 271

 Object Facies Modeling
Adaptive Channel – Cross layer
Body_Index Layering

A Layer tolerance can set up the

maximum number of K-layers that a
Channel (top/base) may cross.
If = 0, only channels whose top K-layer
are in the same layer will be connected.

To get the result to the left:

Layer tolerance > 6 has to be applied

Layer tolerance
The channels honor the stratigraphy of the grid. For channels in
different wells to be connected, their topmost upscaled cells can also
be in a different layer of the grid.
Unless the “Layer tolerance” parameter has been set to 0, the algorithm
first works out the correlations between channel tops. A tolerance value
of 2, for example, means that channel tops can be correlated across a
maximum of 2 k-layers. If the user labeled a channel with body property
“7” at two different wells, but the channel top in each well is in a
different k-layer, then there will be no error or warning results, as long
as the difference is 2 layers or less, and a channel is generated that ties
the two tops crossing the necessary k-layers.

272 • Object Facies Modeling  Property Modeling

 Object Facies Modeling
Adaptive Channel – Well control
The same channel can pass through several selected wells

Channel connectivity control

Channels can also be assigned to particular observations in particular
wells so that the same channel passes through several wells. The
assignment is made by creating a discrete log of an arbitrary ”Body_
index” that spans the same area as the interpretation of facies in the
facies log.

Property Modeling Object Facies Modeling • 273

 Object Facies Modeling
Adaptive Channel – Other output
Using a Body Index property defined in the wells position forces channels to
go through the wells where indexed at the same K-layers
Body_Index Layering
1. Identify a Body_Index property in the Other
output tab of the process dialog.
2. In the Facies bodies tab, select Use Body
Property from Output tab
K=17
K=23-25

The Body index property can be defined

by interactively drawing a discrete log in a
Well section and then upscale it.

Body index
A body index property can be generated interactively in a Well
section as a new discrete log. When the body index property is made,
make sure that the channels defined in the log coincide with the
channel distribution in the facies upscaled property at the correct
K-layers.
Also, make sure that the relationship between the channel and levee is
geologically congruent, because the algorithm always assume that
levee is on each side of the channel at fraction of the thickness.
The Layering property can be generated as a reference by using the
Geometrical modeling process>Zone index method>From all
layers (K) option.

274 • Object Facies Modeling  Property Modeling

 Object Facies Modeling
Adaptive Channel – Geometry Parameters

Specify the layout of the channels:

• Define the Orientation, Amplitude, Wavelength
and Relative Sinuosity

Specify the size of the channel:

• Define the Width and Thickness

Specify the size of the levee:

• Define the Width and Thickness

Adaptive channel - geometry

An important part of any object-based modeling method are geometric
shape and parameters used to represent each facies unit. The adaptive
channel object is represented as channels, or channel/levees.
The algorithm will use well information and trends, as well as
complying with the shape specified by the user for controlling the
distribution of the channels.
Relative Sinuosity
Relative sinuosity indicates how sinuous or tortuous the channel is. It is
For Channel (Section)
defined as the additional channel length compared to the minimum and Levee, make sure the
channel length required to honor the amplitude. The parameter values Width fraction and Width
can be between 0 -1, values greater than 1 are permitted, but results do unit is set properly; if a
distance is set with fraction
not differ significantly from those produced by a value of 1. The value numbers or opposite, it may
must be greater than 0. Increasing the value increases the sinuosity. result in long computation
time as the numbers are too
The most useful range is from 0.1 to 0.4. high or low.

Property Modeling Object Facies Modeling • 275

 Object Facies Modeling
Adaptive Channel – Trends

Trends:
• Vertical (vertical probability distribution-function) and/or
• Areal (horizontal probability distribution-map) or
• Volume (3D probability distribution-volume).

Use channel routing:

Can only be used if either an Areal or Volume
probability is given, because the algorithm only affects
the horizontal distribution of channels and the
weight is a value between 0-1.

Adaptive channel - trends

Three types of trends can be used in the adaptive channel modeling
process: Vertical (1D), Areal (2D) and Volume (3D). If you have
defined vertical probability curves in the Data analysis process, you
can choose to use that instead of a vertical probability/trend function.
The vertical and areal probabilities, or the probability volume, if used,
control the probability of the insertion point of a channel. Additionally,
the areal or the volume probabilities control the routing algorithm. All
input probabilities should be numbers from 0 to 1. Regardless of which
algorithm or probability input is used, the final overall facies proportion
is still controlled primarily by the fraction or number of bodies specified
in the Settings tab (and by the constraints imposed by hard data and
channel geometry). Ideally, the average of the probabilities given as
input should be similar to the requested facies proportion. If it is not,
the algorithm will attempt to honor the facies proportion given in the
Settings tab, and use the input probabilities to modulate the location of
channels according to the selected algorithm.
276 • Object Facies Modeling  Property Modeling
For the Channel routing weight, an entry of 0 is equivalent to not
using the algorithm. An entry of 1 produces the best fit to the given
probabilities, but may adversely affect the channel geometry and
continuity. The default is 0.75 which produces a good fit without
excessively affecting channel shape.

Object Facies Modeling

Adaptive Channel – Background facies

Background facies
When a channel is modeled, Petrel needs to know what to put where no channels are
[Link] background facies can be either undefined, a single facies type or a previously
generated property.

Property Modeling Object Facies Modeling • 277

 Object Facies Modeling
General Object Body Modeling
Object modeling has a general shape/body option to model simple isolated facies
bodies that do not adhere to complex rules like e.g. channels do.

Create a new facies object:

• Select Object modeling as the method.
• Add a new body by clicking the Add a new
body button.
• A new default Ellipse body will be inserted
Other shapes (Ellipse is default)
in the Facies bodies window.

Facies and Fraction:

• Select Facies type from the drop-down
menu.
• Use the upscaled fraction or enter a value.

Object modeling – process

A boolean/object-based simulation algorithm works by randomly
selecting a reference point and then creating a body based on various
criteria, such as fractions, rules for erosion and rules for entire vs.
partial object.
The probability field is based on trend input. In addition, the size, shape
and orientation are drawn from user supplied distribution information.
Objects are inserted into the Facies Bodies list in the process dialog.
The correct order of objects in this list is important. The algorithm
processes each object in the list from top-to-bottom. For example, the
simulation is done for the first object in the list until this has been
completed. Then the next object is processed until complete, and so on,
until all objects in the list have been processed.
If you have objects with dramatically differing fractions or sizes, you
may want to process the object of the greatest size and/or fraction first
and proceed to objects of smaller sizes and fractions next. Of course,
this may be limited by rules (for example, which objects may displace or
erode other objects).
278 • Object Facies Modeling  Property Modeling
 Object Facies Modeling
When can we use General Body Objects?
Body objects is typically used for for isolated facies bodies that does not have a
large width/length ratio and are quite simple in shape:
Facies environments
Some examples from Clastic environments: Eolian Dune
- Eolian dunes (barchans)
- Crevasse splays/lobes
- Alluvial fans
- Sheetflood lobes
- Mouth bars Alluvial Fan
Some examples from Carbonates:
- Patch reefs
- Platform margin isolated reefs
- Reefs talus

Property Modeling Object Facies Modeling • 279

 Object Facies Modeling
General Objects Parameters

Geometry:
• Select Body Shape
• Set the Orientation, Width and
Thickness

Trends:
• Vertical (Function / Data analysis)
• Areal (Probability map)
• Volume (3D property)
• Flow lines (Polygon lines)

Rules:
• Specify whether the facies will
replace other facies (and which one)
or not.

Object modeling - geometry and distribution

All geometrical inputs controlling the body shape (orientation, width
and thickness) are defined by the user. For triangular/uniform
distribution, the values will be stochastically drawn from these
distributions.
Object modeling - probability trends
It is possible to insert Vertical, Areal and Volume probability trends
under the Trends tab when performing Object modeling.
A vertical function (either as a function curve stored in the Input pane
or the Vertical proportion set in the Data analysis process) can be
used together with an areal trend map, but then a Volume (3D property)
is not allowed.
Flow lines can only be used with a vertical trend.
How to generate probability input:
• To make a function, insert a new folder, right-click on the
folder and select Create a new function.
280 • Object Facies Modeling  Property Modeling
 • To make a probability map create some polygon lines, set
values to each line within the polygon and create a surface
using the polygon lines as input.
• A volume probability is essentially a rescaled 3D property
(value range 0 to 1).

Object Facies Modeling

Hierarchical Facies Modeling
Second level facies Models can be constrained by a first level facies models, using
two independent levels of facies models.
An example: A large scale facies model may define regional channel belts and a
smaller scale facies model would model local lithological variations within each large
scale facies type.

Field analog Large scale model Small scale model

Hierarchical modeling
Facies models can be constrained by a second level of facies model in
much the same way as in Petrophysical modeling. A large scale facies
model may define regional channel belts or a transition from shoreface
to deep marine facies. A smaller scale facies model could model local
variations within these regions.
Requirements
Modeling facies at two different scales requires separate discrete well
logs describing the facies at each of these scales. One log will
represent the large scale facies regions, and the second will represent
the individual facies seen in the well.
Property Modeling Object Facies Modeling • 281
 Steps:
1. Define the Hierarchical facies log
2. Click on the Facies button and select an existing populated
large-scale Facies model
3. Select the Facies body to populate with a second level of
facies
4. Define the settings for the different facies

Object Facies Modeling – Exercises

In this exercise, Adaptive channel is used to model the channel-levee
facies in Zone A. The channels in the model will be forced to pass
through several wells. Both channel and isolated object facies are used
to capture the progradation from turbidite lobes to channel levee facies
in Zone B. Both vertical and lateral trends are used to preserve the
spatial facies distribution.
Hierarchical facies modeling where two independent levels of facies
models are used will also be created during this exercise.
For your convenience, an extra project called Property Modeling
2010_M11.pet has been created with all necessary data in case you
have not completed any of the previous modules in the book. If you are
satisfied with your project, continue to work on it.
Adaptive channel modeling
Zone A is interpreted as a channel-levee system and is modeled using
adaptive channels. The channel and levee facies are mostly distributed
along the depositional axis. An adaptive channel object will be inserted
and you will define the settings of this object.
Exercise Steps
1. Open Facies modeling process dialog, and use the existing
property Facies_Adaptive_Model.
2. Deselect the lock icon to open the function panel for Zone A.
3. Select Object modeling (stochastic) as method.
4. In the Background tab>Constant, select to use Shale as
background facies.
5. In the Facies bodies tab, insert an adaptive channel object by
clicking on the Add a new adaptive channel button.

282 • Object Facies Modeling  Property Modeling

 6. Define facies settings as shown in the next figure (remember to
update the Fraction (%) of the channel facies by clicking the
blue arrow button):

7. In the Layout tab, define the orientation and sinuosity of the

channel facies:

Property Modeling Object Facies Modeling • 283

 8. In the Section tab, define the channel dimensions:

9. In the Levee tab, define the levee dimensions:

10. Click on OK to run the model. Check the running Petrel

message log for well conditioning and volume fraction.
11. Check the result using I, J and K filters and compare the result
with the Sequential Indicator Simulation models.
284 • Object Facies Modeling  Property Modeling
Fluvial channels can realistically capture facies architecture. Well data
can control the spatial distribution of facies, in this case to populate
Make sure you have
channel facies along the axis. control, either using Distance
units or Fraction when
setting up the Channel and
Levee width and thickness
Channel and object facies modeling parameters. Accidental
wrong input type may cause
Zone B is a progradational system from turbidite lobes to channel-levee unnecessary long run time.
system. Adaptive channels are used to model the channel-levee facies
in the upper part of the zone, and isolated objects with Deltaic/Alluvial
fan shape are used to model turbidite lobes in the lower part of the
zone. Vertical trends are defined for vertical facies transition, and a
lateral trend map is input to control the lateral distribution of lobe
facies.
Exercise Steps
1. Continue in the Facies modeling process for Facies_
Adaptive_Model (U). Keep the same input parameters of the
adaptive channel facies for Zone B by using the copy
button from the settings of Zone A to Zone B. Select the
Leave zone unchanged button to close the function
panel for Zone A. Go to Zone B, unlock the process and
paste the settings .
2. Remember to update the Fraction (%) of channel facies in
Zone B by using the Blue arrow button to take the fraction
from the upscaled well logs.
3. Go to the Trends sub-tab and use Vertical trends for Channel
facies defined in facies Data analysis, by clicking the Use
vertical function defined in the data analysis button
as shown in the figure:

Property Modeling Object Facies Modeling • 285

 4. Add isolated object facies by clicking on the Add a new body
button. Modify name of body for Lobe and update the
Facies and Fraction as shown in the following figure:

5. In the Geometry tab, input the geometry parameters for the

Lobe facies:

286 • Object Facies Modeling  Property Modeling

 6. In the Trends tab, input Areal and Vertical trends for the Lobe
facies. The Vertical trend is defined as the Vertical
proportion curves during facies Data analysis, while
the Areal probability trend is the Lobe_trend_map stored in
the Input pane>Property Modeling Input folder.

7. In the Rules tab, keep the default settings so that the lobes
will replace all other generated facies types.

Property Modeling Object Facies Modeling • 287

 8. Click OK to run the model and check the Petrel message log
for well conditioning and facies fraction.
9. Check the result through I, J, K-filters and compare with the
conceptual sedimentological model.
10. Open the Data analysis process. In ZoneB, compare the
facies Proportion between the input well data and the
resulting 3D model by selecting the different buttons:
.
In the facies model generated using facies objects, well data are
conditioned, statistics for facies fractions and dimensions are honored,
and most important, the facies architecture from conceptual
interpretation is captured.

Optional Exercise: Adaptive channel modeling with trends

This exercise shows how to model adaptive channels conditioning with
an areal trend (probability map) and volume (seismic attribute).
Exercise Steps
1. Copy - Paste the Facies_Adaptive_Model property, and
rename it ‘Facies_Adaptive_Model_2Dtrend’.
2. Open the Facies modeling process dialog, and use the
existing property ‘Facies_Adaptive_Model_2Dtrend’.
3. Click the Lock button to close the function panel for Zone B.
4. In the function panel for Zone A>Settings tab, change the
Fraction(%) to 16.
5. In the Levee tab, change the Width dimensions:
288 • Object Facies Modeling  Property Modeling
 6. In the Trends tab, define the Areal trend for the Channel
facies (the trend map is stored in the Input pane>Property
Modeling Input folder):

7. Click OK to run the model, display the result and use the I, J,
K-filters to visualize the trend effect over the channel
distribution.
8. Copy - Paste the Facies_Adaptive_Model_2Dtrend
property, and rename it to Facies_Adaptive_Model_
Volume.
9. In the function panel for Zone A>Settings tab, update the
Fraction(%) by clicking the Blue arrow button.
10. In the Trends tab, define the Volume trend by checking on the
option Volume and select the AI_rescaled property. Use
channel routing option and type 0.1 as value for the weight.

Property Modeling Object Facies Modeling • 289

 The AI seismic
attribute has been rescaled
11. Click OK to run the model. Display the result and AI_rescaled
(0-1) to use as 3D probability. properties in separate 3D windows and tile them vertically.
O means no probability Visualize the trend effect over the Channel distribution by using
weighting while 1 is max
weighting. As a compromize, the I, J, K-filters.
a high probability may distort
the channels as they are
elongated bodies by nature, Adaptive channel modeling using body index property
while a given input
probability can have arbitrary This exercise shows how to model adaptive channels that pass through
spatial distribution (like a selected wells. The facies log, Body_Index, has been created and
pixeled 3D volume). If a
continuous trend map of upscaled for this purpose. A channel body indexing log and property will
lateral probability is used, be generated so that the same channel will go through wells DW3 and
the probability can more DW2. The Body_Index upscaled property will be used as input to the
easily be matched.
3D model, and the Body_Index property will guide channels through
the given wells.
Exercise Steps
1. Go to the Templates tab > Discrete property templates >
New Body Index to visualize the Color tab in the settings.
This template was generated to use with the Body Index
property.
2. Go to the Windows pane>Module 11 Object modeling
folder and open the Well section called Well section Body
Index and click the Paint discrete log class button.
3. Create a new facies log by clicking the Create new discrete
Log button which is now active. Select New Body
Index as template for the new log. The new (empty) discrete
log called New Body Index will be added to the Global
well logs folder. Display it in the Well section window.

290 • Object Facies Modeling  Property Modeling

 4. Select the Paint discrete log class button. Right-click
the New Body Index track panel for one of the wells in the
Well section to select which facies type to draw. Left-click to
draw the facies and to move facies boundaries. Continue your
interactive interpretation for other wells on the track panel for
New Body Index.
5. Try to paint 5 different channels according to the figure
below. Use the Layer property (made in Geometrical modeling) There is a pre-made
New_Body_index_
and the red upscaled channels as a guide. As you can see, Complete log and upscaled
some channels should be correlated across (using the same property generated if
needed.
K-layers), while some are independent.

6. When the interactive interpretation is finished, upscale the

New Body Index discrete log by using the Scale up well
logs process.
Property Modeling Object Facies Modeling • 291
 7. Open the Facies modeling process panel, select the existing
property Facies_Body_Index.
8. Toggle off the lock icon to open the function panel for Zone A.
9. Select Object modeling as method.
10. Under the Background tab>Constant, select Shale as the
background facies.
11. In the Facies bodies tab insert an Adaptive channel object by
clicking the Add a new adaptive channel button.
12. Under the Other output tab, select the Body property check
box and select New_Body_Index from the drop-down list, as
shown in the next figure:

13. Back in the Facies bodies tab>Connectivity section, select

the check box Use Body Property from Output tab. Set the
Layer tolerance equal to 0.

Note that the Layer tolerance should be set as a kind of “uncertainty”

parameter, letting Petrel generate connected channels across a certain
amount of K-layers (K-layers are the starting point for the insertion
point of a channel and it builds downwards). With a Layer tolerance of
3, you allow some connection across layers while 0 will not allow any
tolerance, that is, a channel can only be connected if a body index
matches on the exact K-layer between two or more upscaled well logs.

14. Define the facies settings as shown in the next figure.

Remember to update the Fraction (%) of the channel facies.

292 • Object Facies Modeling  Property Modeling

 15. In the Layout tab, define the orientation and sinuosity of the
channel facies:

Property Modeling Object Facies Modeling • 293

 16. In the Section tab, define the channel dimensions:

17. In the Levee tab, define the levee dimensions:

18. In the Trends tab, define the volume trend for the channel
facies:

19. Click OK to run the model. Check the Petrel message log for
well conditioning and volume fraction.
20. Check the result of the Facies_Body_Index property using
the K-filter and compare the result with the Facies_
Adaptive_Model property.
21. Display also the New_Body_Index model and use the
Properties>Filter: Check that one of the channels, for
294 • Object Facies Modeling  Property Modeling
 example, channel 3 pass through the wells DW2 and DW3.
Switch to view the actual Facies_New_Body_Index
property as well.

Optional Exercise: Hierarchical facies modeling

A discrete facies log, called Facies_Hierarchical, has been made to
differentiate between fine grained sand and silt, representing the levee
facies in an already existing facies property. This facies log has been
upscaled and it exists already in the 3D grid called Exercise Model.
Earlier during this exercise, you made a 3D facies model called Facies_
Body_Index. You will use this as the large scale facies model when
the fine grained sand and silt are distributed into the levees in the
Facies_Body_Index property. The method used for distribution of the
two different grain sizes will be Sequential Indicator Simulation.
Exercise Steps
1. Go to the Windows pane>Module 11 Object modeling
folder and open the Well section called Well section
Hierarchical Facies.
Property Modeling Object Facies Modeling • 295
 2. In the Well section, display the wells DW1, DW2 and DW4
together with the Facies_Hierarchical log, Facies_
Hierarchical and Facies_Body_Index properties from the
Exercise Model. Compare the log with the two properties.
3. Open the Facies modeling process dialog. Select the existing
property Facies_Hierarchical.
4. Deselect the Lock button to open the function panel for Zone
A.
5. Click the Facies button to select the already existing Facies_
Body_Index model as a large scale facies model (conditioning
to the Facies_Body_Index model) when modeling the
Hierarchical Facies model.
6. Select the Shale facies and use Assign values as the
Method. Select the Constant option as shown in the next
figure:

7. Use the same Method for Channel facies, but remember to

select Channel as the Constant value.
8. Select Levee as facies and Sequential Indicator
Simulation as Method.
9. Under the Facies tab, insert the Silt and Fine Sand facies by
using the button.
10. Use the same Variogram settings for both facies according to
the next figure.
11. Check the facies fraction under the Fraction sub-tab and
select the values from the Upscaled cells.
296 • Object Facies Modeling  Property Modeling
 12. Click Apply/OK to run the model and check the Petrel
message log for the facies fraction.
13. Check the Discrete statistics under the Facies_
Hierarchical property Settings.
14. Visually check the result using I, J and K filters.

Comment
To guide the distribution along the Levee, a Local varying azimuth can
be used instead of a variogram azimuth. This is an output property from
Adaptive/fluvial facies modeling.

Property Modeling Object Facies Modeling • 297

 Summary
This module dealt with Object modeling. The main difference from Pixel
based modeling is the geometrical definition of channels or general
objects with specific shapes. Adaptive channel modeling was presented
as an alternative for channel modeling where wells are honored.
Hierarchical modeling shows the option for more complexity in Facies
modeling, enabling you to model small-scale variations of facies within
existing facies types.

298 • Object Facies Modeling  Property Modeling

Module 12 - Truncated
Gaussian Simulation

The use of Truncated Gaussian Simulation (TGS) as a method when the

facies show a consistent order in the transition between the different
types, is covered in this module. This algorithm also allows stochastic
distribution of the transitional facies.

Prerequisites
To successfully complete this module, the user must have knowledge of
the following:
• Familiarity with Geostatistics Fundamentals
• Familiarity with Basic Sedimentology
• Petrel Introduction

Learning Objectives
The purpose of this module is to give the participant an introduction to
the Truncated Gaussian Simulation algorithms, in order to use them in
modeling of relevant depositional environments using Petrel. At the
completion of this training, you will be able to:
• Understand the Truncated Gaussian methods available in Petrel
• Use the probabilities, 1D, 2D and 3D trends to describe the
variation of the facies
• Run the Truncated Gaussian Simulation and Truncated
Gaussian with trends algorithms

Property Modeling Truncated Gaussian Simulation • 299

 Lesson

Truncated Gaussian Simulation

Petrel Modeling Techniques for Discrete Properties

Stochastic methods
Pixel Based: Shape decided by Variograms, trends etc.
SIS TGS TGS with trends

Object Based: Facies with defined geometric shapes

General object Fluvial Adaptive Channel

The Truncated Gaussian simulation methods are designed to model

large-scale ordered facies sequences. In addition, one method can also
model progradational and retrogradational sequences, such as those
found in a shoreface or delta front environments.
These methods can also be used in a carbonate environment.

300 • Truncated Gaussian Simulation Property Modeling

 Truncated Gaussian Simulation
Algorithms
Overview of method:
A stochastic method which is able to generate large-scale geometries through
construction of a close facies relationships.
History: TGS w/trends
Known as ’Facies Transition Simulation’ until Petrel
2005.
Modified in Petrel 2007.1 to become ’Truncated
Gaussian with trends’.
In Petrel 2007.1 a new additional method called
’Truncated Gaussian Simulation’ was introduced. TGS
Advantages:
Honors facies transitions / relationships.
Honors well data and global fractions.
Stochastic; providing multiple realizations.

Truncated Gaussian simulation (TGS)

TGS is a standard modeling technique for situations where facies
transitions follow a strict order. Selected facies types use the same
variogram.
TGS in Petrel
The Facies Transition simulation method, available in Petrel 2005
and earlier, was replaced in Petrel 2007.1 by a special version of the
TGS algorithm and was named Truncated Gaussian with trends
(explained later in this module). The Facies Transition simulation
method was a different implementation of the same basic algorithm
and was amended to use the TGS algorithm. The advantage of the
2007.1 version was that the global fraction option was incorporated and
honored.
Truncated Gaussian simulation was another TGS method introduced
in Petrel 2007.1.
Both algorithms are used for situations where the facies transitions
follow a strict order.

Property Modeling Truncated Gaussian Simulation • 301

 Truncated Gaussian Simulation
Theory
Truncated Gaussian Simulation generates realizations of a normalized
Gaussian random field, and truncate the data using multiple thresholds (t)
Process: t3

1
C1
Initial Facies proportions are known from well
C2

0 Proportions
data.
t2
A Gaussian residual field (GRF) using one
Variogram is set for multiple facies t1 C3
The GRF is simulated and subsequently truncated C4
using the thresholds (t1-3) -3 Distribution 3
Results:
The Facies associations (red to purple) are
controlled by the ordering of the Gaussian classes
(C1-4). Red can only be next to yellow and not
purple etc.
Same anisotropy (from one GRF) for all facies

Theory - How does TGS work?

TGS generates realizations of a normalized Gaussian random field,
(normal-scored residual field after trend removal), and truncate the
data using multiple thresholds (from an inverse Gaussian CDF
transform). This will generate several facies where the facies
proportions are determined by the threshold values.
The spatial continuity of the facies distribution is determined by the
variance of the Gaussian field.
Compared to the SIS algorithm, only one Gaussian random field (using
only one Variogram) is simulated for multiple facies; resulting in
faster computation time. On the contrary, the limitation is that it is not
possible to use different anisotropies (variograms) for individual facies.
Various facies types are selected and ordered according to the
ordering of the gaussian classes; for example in the Histogram, the
proportions and thresholds are shown, while the ordering decides
which facies will be modeled adjacent to another facies.
302 • Truncated Gaussian Simulation Property Modeling
In Petrel, the ordering is important when the various facies types are
selected.

Truncated Gaussian Simulation

When can we use TGS?
TGS is typically used to model stochastic unconstrained environments.
Examples could be transitions from Backshore to Shorface

Channel Backshore Lagoon

Upper Shoreface

When can we use the TGS method?

The Truncated Gaussian simulation method is suitable for modeling
areas behind shorface/delta progradation and retrogradation, such as in
lagoonal, barriers and patchy delta front environments. The methods
can also be used in a carbonate environment, related to backshore
patch reefs (sometimes in conjunction with simple shape object
modeling).

Property Modeling Truncated Gaussian Simulation • 303

 Truncated Gaussian Simulation
Facies & Fractions

The Facies association can be

ordered and each facies
proportion must be specified:

The order of facies is important

when using this method (move
facies up/down once selected)

Specify the Global fraction from

Upscaled cells, Well data, Manually
or by a Property trend

Facies Associations
Truncated Gaussian simulation is a method that is used when the
facies show a consistent order in the transition between the different
types (often seen in carbonates or shoreface facies). The ordering of the
facies is key in this case – in Petrel, you can select which facies to
model (from the left field on the Facies tab), drop them over using the
blue arrow. Now, the facies types can be moved up or down depending
on which facies they should be modeled adjacent to.
Global Fraction/Facies proportions
This method can take various probabilities: Vertical, horizontal or 3D
trends to describe the variation of the facies.

304 • Truncated Gaussian Simulation Property Modeling

 Truncated Gaussian Simulation
Variograms & smoothing
The method simulates only one
Gaussian random field:

In the Variogram tab, each facies

must follow the same variogram
Gaussian is the recommended
variogram model type

In the Settings tab, one can

smooth result between the facies
borders/ transition (in terms of cells
distance I,J and K.

Variograms
Only one Gaussian random field is simulated even for multiple facies
types, so each of the facies must also follow the same variogram.
Smooth
If the Variogram settings are small or the nugget high (increased
applied variance), the result will be more arbitrary/pixeled. To “gather” When using
the facies more, a smoothing option is applied in the I,J and K smoothing, the initial
fractions used will not be
directions. This will smooth the facies boundaries. honored as well as with no
smoothing.

Property Modeling Truncated Gaussian Simulation • 305

 Truncated Gaussian with trends
TGS algorithm

Pixel-based method used for modeling ordered facies transitions

Facies boundary type can be linear or curved
Interactive interface for facies boundary editing
Interfingering effect at facies boundaries
Clinoform for facies interfingering

Truncated Gaussian with Trends

Truncated Gaussian with trends allows a stochastic distribution of
the facies based on a given transition between facies and a trend
direction. The trend shape and direction is set interactively in the dialog
window and a range is given for the variogram.
The trend boundaries between the various facies can be moved
interactively. These trends are converted into probabilities and the
model is run using the standard Truncated Gaussian technique. The user
can specify global fractions for each of the facies.

306 • Truncated Gaussian Simulation Property Modeling

 Truncated Gaussian with trends
When can we use TGS with trends?
TGS w/trends is typically used for facies in Carbonate build-outs and Delta fronts
where large-scale progradation / retrogradation is prominent
Carbonates

Clastics Type Log

Channel Backshore Lagoon
Upper Shoreface
Lower Shoreface
Platform

When can we use the Truncated Gaussian with Trend Method?

The Truncated Gaussian with trend can model large scale ordered
facies progradation and retrogradation, such as in a shoreface or delta
front environment involving , for example, progradational fluvial
sequences.
These methods can also be used in a carbonate environment, with
prograding and aggrading reef structures.

Property Modeling Truncated Gaussian Simulation • 307

 Truncated Gaussian with trends
Principle

0.4
0.6
0.81.0
1.21.4
+
1.6 1.8 2.0 2.2 2.4 2.6

Deterministic trend based on well Residual field based on variogram

data and interpretation (to keep the and well data (for interfingering effect)
facies ordering and boundaries)

The final model by re-truncating the combined field

Principle
The deterministic trend is based on the point or line source location,
orientation, and the position and angle of the facies boundaries. The
facies boundaries can also be edited.
The residual field is computed between the deterministic trend and
the upscaled cell values. For example, if the deterministic trend value is
equal to upscaled facies value then the residual value is equal to 0.0.
If the upscaled cell value is different from the trend value, there is some
residual value to distribute into the cell between well locations.
Residual values are distributed based on interfingering parameters (that
is, a variogram and a variance value that describes the correlation of
residual values). The variance is used when calculating the residual
values in the upscaled wells. The square root of the variance is used as
the standard deviation in the Gaussian algorithm.

308 • Truncated Gaussian Simulation Property Modeling

 Truncated Gaussian with trends
Process
1. Facies 2. Transition trend 3. At each well: Simulation value
transition zones surface Log = 2, Trend = 1.7 -> Residual = -0.3
Map with residuals at well position
-0.3
0.25
-0.11
0.18
0.31

4. Residual property 5. Add residual field to transition field

Process steps:
1. The user defines the facies transition zones as a 3D trend
(integers)
2. The transition trend surface (real numbers) calculated from the
facies zones
3. At each well: Simulation value = trend surface - log facies
4. Calculation of residual property using Gaussian simulation
technique
5. Add residual field to transition field

Property Modeling Truncated Gaussian Simulation • 309

 Truncated Gaussian with trends
Facies

Facies transition:
- Facies must be selected
in correct order
- Facies can be moved to
fit ‘adjacent’ to correct
facies types
Facies fraction:
- Specify a Global fraction
for each facies type
- Geometry trend adheres
to Geometry tab settings
(next slide)

Truncated Gaussian with trends

Geometry (trend) and Variogram

Geometry and build-up style:

- Geometry decides the 3D trend
- Facies transition lines can be
edited

Variogram Ranges and Variance:

- Variogram decides the spatial
continuity
- Variance decides the degree of
interfingering along the
transition (trend) lines

310 • Truncated Gaussian Simulation Property Modeling

Geometry tab (Trend model)
The method involves first choosing which facies codes to be included in
the sequence and in what order, then defining a trend along which the
facies codes are expected to change. Residuals between the trend and
the well logs are then distributed using the Truncated Gaussian
algorithm and the defined variogram.
If the trend does not match the input data, the result may be noisy.
Variogram & Variance
The variogram is set once for all facies types. Increasing the range will
apply more correlation and bundling of each facies.
If the Variance is also increased, the variogram together with the
increased variance will create an elongated/connected interfingering
effect.

Truncated Gaussian with trends

Geometry tab – Aggradation angle
Aggradation angle is the angle between facies boundary and grid layer

Depositional architecture: Galloway 1989

2D trend model
3D property model

Property Modeling Truncated Gaussian Simulation • 311

 Aggradation angle
The aggradation angle can be varied from progradation (positive
shoreline trajectory; building outwards) to retrogradation (negative
shoreline trajectory; backstepping). With an angle of 90 deg, it is
considered aggrading.
You can use the same angle for all facies transition lines, but you can
also use different angles.
Transition lines – explanation:
Each facies has its own transition lines, a solid (top), and a stippled
(bottom) line according to the facies color.
The arrow indicates the movement of the shoreline trajectory. If the
model is progradational, the bottom transition line can only be pulled
backwards in the direction away from the arrow. If retrogradational, the
opposite applies.

Truncated Gaussian with trends

Geometry tab – Edit transition lines
Edit lines: Normally used together with interfingering (variance)

Note: Do not use too much Variance, as it will clutter the more detailed facies
boundary editing.

312 • Truncated Gaussian Simulation Property Modeling

Edit transition lines
Once an overall large-scale trend model is built, each transition line can
be edited to create curves (Spline) or straight lines. This way, a small
scale adjustment to each upscaled well log can be made and a more
detailed transition model can be made based on the standard Line or
Point source.

Truncated Gaussian with trends

Variance and Variogram Range – Controls interfingering

Low Range High Range Low Range High Range

‘connected’ ‘connected and ‘high frequency’ ‘low frequency’

interfingering noisy’ interfingering interfingering interfingering

Do not use too much

Variance, as it will clutter the
more detailed facies
boundary editing.

Property Modeling Truncated Gaussian Simulation • 313

 Truncated Gaussian Simulation - Exercises
The Truncated Gaussian Simulation methods are designed to model
large scale ordered facies progradation and retrogradation such as in a
shore face or delta front environment. The underlying algorithm for
these methods is the Truncated Gaussian simulation (TGSim).
For your convenience, an extra project called Property Modeling
2010_M12.pet has been created with all of the necessary data in case
you have not completed any of the previous modules in the book. If you
are satisfied with your project, continue to work on that.

Modeling a transitional depositional environment using

the “Truncated Gaussian simulation” method
Truncated Gaussian simulation uses a simple setup where the ordering
of facies is important. The selection of the facies order and their
fractions will determine the final result.
Exercise Steps
1. In the Exercise Model>Properties folder, an upscaled facies
log named ‘TGS (U)’ already exists. This is a facies log created
just for the purpose of this exercise (for both ZoneA and
ZoneB).
2. Open the Facies modeling process, select existing the
property ‘TGS (U)’.
3. Deselect the Lock button to open the function panel for Zone
A.
4. Select Truncated Gaussian simulation as the modeling
method.
5. Under the Facies tab, insert the Proximal, Middle 1, Middle
2 and Distal facies by using the button.
6. Remember, it is important to keep the four facies in the
correct order. Therefore, if you need to move the Middle 2
facies upwards, use the Move Selected item up button.
7. Check the facies fraction under the Fraction tab and keep the
default Manual option, with the values calculated from the
upscaled cells. Remember to select the Trust fraction or
trends check box:

314 • Truncated Gaussian Simulation Property Modeling

 8. Under the Variogram tab, specify the variogram settings
according to the next figure:

Property Modeling Truncated Gaussian Simulation • 315

 9. Under the Settings tab, select to smooth the result according
to the settings in the next figure:

10. Run the model and check the result against the well data using
the I, J and K filters.

Modeling a transitional depositional environment using

the ‘Truncated Gaussian with trends’ method
A conceptual sedimentological model is required to define the facies
transitional pattern including source type, depositional direction and
progradational angle. These parameters can be interactively defined
using a graphic interface or they can be typed in. Interfingering effects
can be added to capture the irregularity of the facies boundary and
honor the localized variation of the facies boundary at the well
locations. The interfingering effect can be defined through a variance
parameter.
Exercise Steps
1. In the Exercise Model>Properties folder, an upscaled facies
log named ‘TGS_with_trends (U)’ already exists. This is a
facies log created just for the purpose of this exercise (for both
ZoneA and ZoneB).
316 • Truncated Gaussian Simulation Property Modeling
 2. Open the Facies modeling process, and select the ‘TGS_
with_trends (U)’ property.
3. Deselect the Lock button to open the function panel for Zone
A.
4. Select Truncated Gaussian with trends as the modeling
method.
5. Under the Facies tab, insert the Proximal, Middle and
Distal facies by using the button. Remember to keep the
three facies in the correct order.
6. In the Fraction sub-tab, use the Geometry trend option to
define the Global fraction as shown in the next figure.
Remember to select the Trust fraction or trends check box:

7. Under the Geometry tab, define the facies boundaries using

the graphic interfaces and type in boxes. Make sure that the
check boxes for ‘Line source’, ‘Progradational’ and ‘Use
different’ (aggradation angle) are selected.

Property Modeling Truncated Gaussian Simulation • 317

 8. Well location and upscaled facies logs can be
posted in the map view to help defining the facies boundaries.
9. To check how well the facies boundary fit to a well, click the

well symbol black button , and the well facies

profile will be displayed on the right side of the dialog.

10. Edit the lines that define the lower (stippled) and upper (solid)
transition zone, by positioning on the lines and dragging them
interactively.
11. Define the Variogram settings under the Variogram tab. The
amount of irregularity and the continuity can be controlled by
the Variogram type, Nugget and Ranges.
12. The degree of interfingering is highly affected by change in
Variance; try a variance of 0.00001 (approaches trend) and
0.01 (giving a better scattering/interfingering effect).

318 • Truncated Gaussian Simulation Property Modeling

 13. Click Apply to run the model and check the result against the
conceptual model and well data using the I, J and K filters.
14. Go to the Geometry tab, select the Edit lines check box and
add points to the facies lines. Move them to create undulating
lines. Click OK to rerun the model and check the result.

Property Modeling Truncated Gaussian Simulation • 319

 Summary
In this module the Facies modeling techniques were presented,
producing results of distinctly ordered sequences. This can be done
with or without trends. Because Truncated Gaussian Simulation is a
stochastic algorithm, many options are available, such as multiple
realizations and trends.

320 • Truncated Gaussian Simulation Property Modeling

Module 13 - Petrophysical
Modeling

The key concepts in Petrophysical modeling are covered in this module

including data analysis to prepare input data, how to use different
trends and how to identify the spatial variation for each petrophysical
parameter.

Prerequisites
To successfully complete this module, the user must have knowledge of
the following:
• Familiarity with Geostatistical Fundamentals
• Familiarity with Petrophysics Fundamentals
• Familiarity with Reservoir Modeling
• Good general knowledge of Petrel interface and processes

Learning Objectives
The purpose of this topic is to give the participant a basic
understanding of the available Petrophysical modeling and Data
analysis tools in Petrel.

At the completion of this module, you will be able to:

• Prepare data and its subsequent quality control using the Data
analysis tools.
• Perform statistical analysis for continuous data, including how
to transform data and how to create variograms from data to
perform Petrophysical modeling.

Property Modeling Petrophysical Modeling • 321

 Lesson

Petrophysical Modeling
Overview

Key Issues
Different petrophysical property
distributions in different facies
Various trends
Spatial variation for each
petrophysical parameter
Correlation between parameters

Identify petrophysical features critical to production

Petrophysical Modeling - Overview

The fact that petrophysical properties vary by facies is reason for doing
facies-biased log curve upscaling and perform facies modeling first.
In the presence of facies as a conditioning parameter, petrophysical
properties are subjected to variogram analysis, and trends are identified
per facies.
The correlation between parameters may be porosity vs. acoustic
impedance or permeability vs. porosity, and can also be used to
condition Petrophysical modeling.

322 • Petrophysical Modeling Property Modeling

 Petrophysical Modeling
Petrel Modeling Techniques for Continuous Properties
Methods used in the course:
Deterministic: One single output
Kriging

Stochastic: Multiple equally probable outputs

SGS GRFS

Deterministic techniques are typically used when dense data is

available (for example, many wells, wells and seismic). The
deterministic methods yield a single estimated result (that is, they do
not produce multiple realizations).
Kriging is a family of estimation techniques in a probabilistic
framework. The principal goal of Kriging is to produce the best result in
terms of local accuracy.
Stochastic techniques are often used in conditions where sparse
data is present. These methods produce a possible result and can be
used to generate multiple equally probable realizations.
The petrophysical modeling approach in Petrel can be deterministic,
stochastic or a combination of both. Each zone in the model is given a
specific setting and filter sensitivity can be applied to each process. You
can filter on facies, values, index, zones and/or segments.
This course concentrates on the Sequential Gaussian Simulation (SGS),
Gaussian Random Function Simulation (GRFS) and Kriging methods. We
will not cover every algorithm available in Petrel in this course.

Property Modeling Petrophysical Modeling • 323

 Petrophysical Modeling
Modeling Inputs

Well data: upscaled/blocked well logs

Distribution: histogram
Variogram (spatial model):
- Direction, model type, nugget and sill
- Correlation lengths in 3 directions (range)
Facies model: Conditioning
Spatial trends: From seismic /analogs etc.
Secondary parameter: with a correlation

Modeling input
Blocked well logs is the same as upscaled property cell data. This is
normally the main input to Petrophysical modeling.
Examples of spatial trends: Porosity decreases with depth or a regional
porosity decreasing trend basin-wards.
Examples of secondary parameters with a correlation: permeability vs.
porosity or porosity vs. acoustic impedance.

324 • Petrophysical Modeling Property Modeling

Petrophysical Modeling
Statistical Data Analysis – Continuous Properties

Data transformation: data distribution and spatial trends

Variogram analysis: spatial variation
Correlation: relationship between parameters
By interval (zone) and by facies: maintain heterogeneity and
difference

Data analysis - Petrophysical modeling

Petrophysical data analysis is a process of quality controlling the data,
exploring the data, and preparing inputs for Petrophysical modeling. The
analysis can be done by selecting the Property, Zone and Facies model.
Petrel provides statistical continuous Data analysis functionality
including data transformations and variogram analysis.
Data Transformation examples
• Outliers in the input data can be eliminated via input
truncation.
• Trends can be identified and removed (e.g., vertical trends,
areal trends or 3D trends based on correlated secondary data).
• Output properties can be truncated to force realistic values
(e.g., eliminate negative NtG or porosity values).
• Gaussian simulation requires a normal distribution shape with
a given mean and standard deviation, so this is the purpose of
a normal score transformation.
Variogram Analysis
This is conducted using facies constraints and implies variograms are
computed and modeled for each facies.
Property Modeling Petrophysical Modeling • 325
 Correlation
Function windows and trend transformation can be used to identify
which second attributes might be useful.
By Sequence and Facies
Data analysis can be performed per Zone and per Facies to maintain the
distinct character of petrophysical property per interval and facies.

Petrophysical Modeling
Data analysis process – Distribution (by individual Facies)
Histogram for different Facies:
Channel
Is the histogram natural or does it need to be edited? Phi =0.22

Shale
Phi = 0.02

Input distribution
Lobe
for one facies type Phi = 0.10

Data Analysis - Input Distribution

Users should examine the histogram distributions of a given
petrophysical property (for each facies) if these have already been
identified. Such investigation often reveals outliers; these can be
truncated using available transformations (see next slides).
Notice how the distribution of porosity is dramatically different for each
facies: Some distributions contain spikes, some are multimodal, others
are somewhat skewed and appear to have a lognormal distribution. 
Consequently, you need to examine and analyze the histograms of
petrophysical data independently for each facies.
326 • Petrophysical Modeling Property Modeling
 Petrophysical Modeling
What is a Transformation?
Transformation is the preparation of a real data set into an internal data set.
Several transformations can be run in sequence. Before a simulation algorithm is
run, a final Normal Score transformation is used (standard normal distribution:
Mean =0, [Link]=1.)
Normal Score
transformation Back- transformation
Input Output
distribution µ=0,s=1
distribution
(well logs) (3D property)

Back-transformation will automatically be performed in the reverse order of

the initial transformations to preserve the spatial trends and original data distribution
in the resulting property.

Transformations
Transformations available in the Data analysis process in Petrel
include:
Input truncation - Truncate the input data.
Output truncation - Truncate the output data after the modeling
algorithm is run.
Logarithmic - Logarithmic transformation.
Cox-Box - Removes skewness from the data.
1D trend - Investigate and temporary remove linear trends.
2D trend - Investigate aerial trends using a surface.
3D trend - Investigate spatial trends using a property and a correlation
Shift scale - Shift mean and scale the standard deviation of the data.
Normal Score - Transform the data to a standard normal distribution
(Mean=0, Std. dev=1).
Property Modeling Petrophysical Modeling • 327
 Petrophysical Modeling
Data analysis process – Transformation (Distribution)
Truncate distribution range

Anomalous Gamma
distribution for porosity
in Sand (cementation)

Trucates input distribution to delete

unrepresentative values or push
them to next bin

The Output truncation has no

Histogram representation as it is
performed after simulation as a
limit on the CDF curve using a
max/min (not a physical cut of
Truncates the output of a realization input data)
on back-transformation of data, to
get values in a desired range

Truncating distribution range

These transformations help to remove initial data that are not
representative, or help limiting the modeling output distribution:
Input truncation - Truncates the input distribution to remove data not
to be represented in the final output. These anomalous/
unrepresentative values are removed by a Minimum and Maximum
cutoff parameter.
• Force outside values to the boundary will include low values at
the specified minimum/maximum. Typically, this could be based on
Calculator cutoffs on a log where a zero-value or ‘low-cut’ value is
used but where there should be undefined/not investigated values.
• Outside values undefined will ignore values above or below the
maximum and minimum respectively. For example a Sandstone
with some shale or cementation not representing the ’true’ porosity
can be taken out completely from the input data.

328 • Petrophysical Modeling Property Modeling

Output truncation - is the last step of the back-transformation of the
data. Therefore it has no effect on the histogram within the
Transformation tab. The Output truncation will be performed on the
generated 3D property to ensure that the it does not get values outside
the desired range.
Example: If you are modeling porosity in clean sand, you may expect
high porosity values; set the output truncation to make sure you do not
create too low or too high values

Petrophysical Modeling
Data analysis process – Transformation (Shape and Scale)
Change distribution shape

Gamma distribution
Permeability

No input parameters required

Lognormal distribution
of Permeability
Lobe facies (Zone B)

Removes skewness of data by

a Lambda factor to approx.
Normal distribution

Distribution shape
Some transformations can change the distribution shape by using
logarithmic transforms or a power law exponent value to change the
shape of a skewed distribution to approximate more a normal
distribution.
Logarithmic - is a useful parameter for changing a highly skewed
shape of a distribution. This will normally be done as a final step of
normal score, but then the difference from a gamma or lognormal
distribution shape to the symmetric ’bell-shape’ of the normal
Property Modeling Petrophysical Modeling • 329
 distribution is large. For example, a log transform should be performed.
Typical properties to perform on are Permeabilities.
Cox Box - removes the skewness from the distribution. The factor
Lambda expresses the degree of skewness and can be entered
manually or estimated automatically. It is a Power factor that changes
the shape of the distribution curve. Lambda=1 is approximately the
same as the original distribution.
Pressing the Estimate button automatically estimates Lambda
beginning in the range -16 to 16 and iterating down to an accuracy of
0.1, aimed at maximizing the likelihood function of a normal
distribution.
If Cox-Box transformation is applied, you will usually have to apply the
Shift Scale transformation afterwards.

Petrophysical Modeling
Data analysis process – Transformation (Distribution Shift/Scale/Shape)
Change distribution range & scale

Mean por = 0.21 Mean por = 0.11

Shift data by mean and scale by

standard deviation – done after New target mean = Input mean shift - real mean / [Link]
spatial transformations

Distribution curve
can be edited
with caution

Forces any distribution shape to

standard normal (m=0, std=1)

Moving mean and changing scale of distribution

Shift scale - is used to shift the data to a target or expected mean,
and scale the curve using the standard deviation. It should usually be
330 • Petrophysical Modeling Property Modeling
applied after spatial transformations like Cox-Box, Logarithmic and
Trend transformations. Unlike Normal Score, it will not change the
shape of the distribution.
Example: If a Porosity range from a limited set of wells are upscaled
and is not representative of the entire grid to be modeled, the overall
mean can be shifted up or down. For example, if the wells target only
sands, the mean should be shifted down. To keep the distribution range
(no scaling) set [Link] to 1 and only shift the mean down. To shift the
porosity down 0.1 (10%), apply a mean (shift value – not final value).
Note that from the formulae a positive shift value shifts the mean
down.
Normal Score - will force any distribution to a standard normal
distribution (mean = 0, standard deviation = 1) with most of the samples
close to the Mean value, while few samples are extreme/outliers.
Normal Score should always be the last transformation performed on
your data.
Normal Score transformations should be used with caution if you have
limited input data, as they will force the distribution of your property to
exactly match the distribution of your input, and the histogram can be
unrepresentative. This can be avoided by using a Shift scale, Cox-Box or
Logarithmic transformations, or by defining the histogram curve
manually (see slide). Shifting the curve manually is more of an expert
exercise and it is difficult to maintain/reproduce as it is fully manual.
Shift scale and other transforms are better at keeping ‘tidy’ parameters.
Options for Normal Score transformation:
• Use values - Transformation purely performed on your own input /
transformed data. Typically used with a large number of data
points with a reasonable spread (i.e. a smooth histogram)
• Define curve - This option will allow you to edit the curve for the
transformation, i.e. allow you to make a model by using a different
distribution from that in your input data. Use it when the input data
is sparse and the histogram is uneven. If many of the data points
have the same value, spikes will occur in the histogram. Checking
the Despike option will ensure that you get a normal distribution
even with many equal values.

Property Modeling Petrophysical Modeling • 331

 Petrophysical Modeling
Data analysis process – Transformations (Trends)

Trends in the data must be accounted for and removed to honor the concept of
Stationarity. Trends could be 1D (vertical trend related to compaction with depth),
2D (lateral facies variations) or 3D (correlated trends from another smoothly varying 3D
property; e.g. Acoustic Impedance).

Trend process: Porosity - Real Porosity - Residuals

Non-stationary Stationary
Is there a trend in the data? Investigate.
If a trend is found, apply a Trend function. 0.1 -0.1
Trend data is temporarily removed Trend removal
(decomposed into a smooth trend and a +0.2
0.1 0.4
residual). -0.1
The residual (random component) is
0.2 0
simulated.
After simulation, the trend is added back
(during back-transformation). Trendline = 0.2 No Trend

Spatial trends
Spatial trend can be investigated; removal allows residual values to be
used in modeling. The assumption is that residual values have a higher
likelihood of exhibiting a normal distribution when a strong trend is
present. In addition, de-trending the data more easily allows the spatial
structure of the variogram to be seen and modeled.
According to the requirements of GSLIB algorithms (e.g. SGS), the data
have to be transformed in order to become stationary before the
modeling process. This means that the spatial trend needs to be
removed and the distribution needs to be transformed into standard
normal distribution so that the mean is 0 and standard deviation is 1.
Trend options in Petrel
1D trends are normally related to porosity values; these will for the
same lithology decrease with depth due to compaction and possible
diagenesis.

332 • Petrophysical Modeling Property Modeling

2D trends can be related to lateral variation of lithology (surfaces in
Petrel).
3D trends are related to full properties that exhibit a trend. This can be
used to avoid bias in modeling. The 3D trend should be smooth.

Petrophysical Modeling
How to do a 1D Trend Transformation in Petrel
How to use 1D Trend in the Data Anlysis process:
Drop in a 1D trend using Blue arrow and select Show: Input.
Fit the Trend function (blue) to the Regression curve (gray) or move it
Select Show: Output to view the distribution of the residual to be modeled
Enable/disable selected
transformations

The blue line is the APPLIED trend

and will remove the trend
unless disabled
Trend removal

Original Distribution Distribution after trend removal

Property Modeling Petrophysical Modeling • 333

 Petrophysical Modeling
Example of a Transformation sequence

Before modeling, Petrel will perform the

following transformations:
• Truncate the input distribution (i.e. eliminate
outliers)
• Remove the 1D trend (vertical compaction)
• Normal score the data (mean of 0, std of 1)

Perform modeling based on the transformed

data set. Then back-transform the data:
• Remove the normal score transform
• Add the 1D trend that was removed
• Truncate the output distribution (using set Max
and Min. values)

Example of a transformations sequence:

1. Input truncation (Set truncation; values outside min and max
are set to undefined or pushed into next bin)
2. 1D vertical trend removal of a well defined trend (perform the
trend fit in Simbox space mode, as you will usually remove
trends in Simbox mode).
3. Normal score transformation (i.e. transform the data set into a
distribution with mean of 0, std of 1)
When you run Petrophysical modeling, the following will occur:
1. The 1D trend creates residual values that are normal score
transformed
2. The algorithm will distribute the normal scored residual values
into the cells
3. The result will be back-transformed according to the normal
score parameters
4. The 1D trend will be added back.
5. The data will be truncated according to Output truncation min
and max values.

334 • Petrophysical Modeling Property Modeling

A 3D property is generated, where the output distribution honors the
input/output distribution, and contains the same 1D trend as observed
in the input data.

Petrophysical Modeling
Data analysis process – Variogram Analysis

The variogram is calculated on

transformed data. It measures
variability with distance.
Calculated in 3 directions:
– Horizontal Major
– Horizontal Minor
– Vertical

A search cone must be set to

capture data within lags

Points = Sample variogram

Line = Regression line 1 Nh 2

Line = Model Variogram g (h ) =

2Nh
∑ ((Φ
i =1
(i +h ) ) − (Φ ))
i

Histogram = Number of pairs

Variograms
Transformation - Data should be transformed (and always with
Normal Score) before modeling the variogram, otherwise the variogram
sample points are not understood by the process. However, if there are
trends in the data, the sill/range is not easily determined as the sample
points will not reach a plateau. These must be removed.
Sample/Experimental variogram - Are computed and represents a
measure of average variation of data points (squared difference) at a
given separation distance (lag). The left hand Y-axis indicates the
semi-variance values.
Histogram bars - The underlying histogram bars represent the number
of pairs used for creating the average value of each sample point. The
Property Modeling Petrophysical Modeling • 335
 number of pairs are indicated in the right hand Y-axis. Notice that the
number of pairs decrease with increased lag distance.
Variogram model - The line (shown in blue) is the variogram model
and is the result of a curve fitting exercise based on the experimental
variogram points.
A quick review of the three variogram “model types” available in Petrel: 
• Spherical: Simplest and has a linear behavior at shorter
distances with a sharp transition to a flat sill. Spherical
variograms are most robust and stable in terms of the Kriging-
equation system to be solved. Influence of the data points are
limited by the Range. 
• Exponential: Has a steep behavior at shorter distances with
an asymptotic approach to the sill at longer distances. 
• Gaussian: Reserved for phenomena that show high degree of
continuity at short distances and then begin to transition to
more of an exponential behavior at longer distances. Rarely
used for porosity, should be avoided for discrete data types,
and used with caution for permeability when justified.
Nugget - The nugget and model type must be the same for all three
directions. Start with the vertical variogram since this is usually better
defined, and establish the nugget based on this. The nugget represents
the variance at a range below the lag distance, or the smallest
separation between points.

336 • Petrophysical Modeling Property Modeling

 Petrophysical Modeling
Variogram – Search Cone

Suggested Lag distance:

Vertical = cell thickness

= log sampling
Y axis

Angle=60o

Lateral = well spacing

X axis
Variogram search cone
If the search cone is set up correctly, the sample variogram will
automatically plot as well as the data quality will allow.
This slide gives a good explanation of the concepts of direction, search
radius, angular tolerance, bandwidth, lag, and lag tolerance. Together
these parameters define lag “bins”. Data pairs are identified based on a
lag bin methodology. All data pairs to the same base lag contribute to
the experimental variogram value for that respective lag distance.  
Bandwidth: A distance cutoff used to prevent the lag bin search area
from becoming too large (i.e., wide) at lag distances far from point of
origin. 
Angle Tolerance: It would be too restrictive to expect all pairs in a
given direction to lie along the exact line representing the selected
direction. This tolerance provides some leeway so that data pairs can
be identified, approximating a given direction without being too
restrictive.
Property Modeling Petrophysical Modeling • 337
 Lag tolerance: Distance +/- the lag spacing within which data will be
considered belonging to a given lag. Typically this tolerance defaults to
½ the lag spacing. In short, we cannot afford to loose data pairs as
most analysis on upscaled well log data struggle having enough data to
start with. 

Short Variogram recap
• Make sure the Search radius is appropriate. For vertical
wells, it is often too big (due to Zone and/or Facies filtering).
• Focus on the first part of the variogram. It is better to fit the
variogram model curve to this part than to approximate a
best-fit to all points, like the default regression curve does
(grey colored).
• Using a larger Lag tolerance to get more pairs and a
smoother variogram.
• Start with a simple variogram (a Tolerance angle of 90
degrees simply means an omnidirectional variogram) before
plunging into directional variograms.

Petrophysical Modeling
Variogram Modeling Process

Vertical variogram model

– plenty of data
– easily estimated
Fit the model variogram to the experimental variogram
– spherical, gaussian, exponential
Horizontal variogram
– usually limited data, hard to make a good variogram
– Can be:
• implied from geology knowledge
• can be derived from correlated data source

Variogram modeling process

The three first points in the slide have been discussed earlier and we
will now concentrate on the last point:

338 • Petrophysical Modeling Property Modeling

Often, well data is too sparse to facilitate variogram modeling in the
horizontal direction. For example, a single exploration well will only
For carbonate
support vertical variogram analysis as there are no pairs available in the reservoirs there is a lot of
horizontal direction. data in the horizontal
direction and less data in the
In this case, it is common to turn to a correlated secondary source of vertical direction due to
data. Given a reasonable correlation, one can justify the horizontal many horizontal wells. In this
situation it is easier to create
analysis on the secondary data. This data is used as a proxy or a horizontal, rather than a
substitute for the purpose of interpreting the direction, major and minor vertical, variogram.
range values.

Petrophysical Modeling
Variogram – Using Secondary data as lateral constraint

Variograms - Use of secondary data

Variograms gives local constraints on the spatial distribution of data. To
generate a good variogram in all directions is critical to get an
understanding of the spatial variability of our data and how correlated
they are. When we use any variogram-related algorithm, we should
have a pretty good idea of the parameters of the Variogram ellipsoid.
Remember that the variogram varies with different Facies, Lithologies
and Rock types:
Property Modeling Petrophysical Modeling • 339
 Initially, we find the vertical variogram range as this is easy to find
using densely sampled vertical well logs. Also, the nugget is good to
find (and we only need one nugget value for all 3 directions).
If there is a lot of wells present and/or there are horizontal wells, we
may attempt to find horizontal variogram ranges. Due to lack of data
this may be too coarse and even futile.
However, if any correlated data source has been found (like a
resampled acoustic Impedance cube) and it exist in every cell location
to be populated, it can be used as a secondary variable when doing
stochastic modeling.
Since it is used in the actual modeling, we may also derive the
remaining Variogram parameters; major direction (if anisotropy), major
and minor range. The Major direction can be found from a Variogram
map, and the Major range can be found from the sample variogram in
the direction indicated by the map. The Minor range can be found 90
deg to the Major direction. If the reservoir is homogenous and isotropic,
the major and minor ranges are equal.

Data Analysis – Exercises

Petrophysical Data analysis is about quality checking the input and
modeled data, exploring the data and preparing inputs for Petrophysical
modeling. The two major parts of Petrophysical Data analysis in Petrel
is transformation and variogram analysis.
For your convenience, an extra project called Property Modeling
2010_M13.pet has been created with all necessary data in case you
have not completed any of the previous modules in the book. If you are
satisfied with your project, continue to work on it.

Data Transformation
Under the Transformation tab in the Data analysis process, you can
view the histograms to check the distribution of the reservoir property.
You can check the difference in the petrophysical properties between
teh facies using histograms generated for different facies. In cases
where little or no petrophysical samples are available to generate a
representative histogram, a user-defined distribution can be provided
for later petrophysical modeling.
340 • Petrophysical Modeling Property Modeling
According to the requirements of the simulation algorithms (SGS or
GRFS), data has to be stationary (transformed into standard normal
distribution) before the modeling process. This means that the trends
needs to be removed and the distribution needs to be transformed into
a mean of 0 and standard deviation of 1.
The simulation result will automatically be back-transformed in the
reverse order as set in the Data analysis process. By doing so, the
spatial trend observed in the input will be added back and preserved.
The original distribution will also be honored via the CDF-curve used for
back-transformation.

Applying Normal Score Transformation

Exercise Steps
1. Make sure Exercise Model grid is selected. Open the Data
analysis process from Processes pane>Property modeling
and select ‘PHI [U]’ property.
2. Notice that Zone A is already selected. Click on the unlock
button to open the process and the button to
condition to facies and select the ‘Facies_Adaptive_Model’.
3. Note that in the Transformation tab, Normal score is
already selected (default in Petrel).

4. In the Transformation tab, review the Histogram window in

the left corner for all the different facies by selecting the facies
code. Check the parameters (Min, Max, Mean and Std).
5. In the Show option, change between the Input (original data)
and Output (after selected transformation) histograms.
Observe the difference between both of the histograms shapes
and parameters.
6. Click Apply to save the changes.

Property Modeling Petrophysical Modeling • 341

 The values in the Output histogram are distributed in Normal score.
After the transformation, note the changes in the values going from the
original porosity values to the Normal Score values. For example:
Original: Min. 0, Max. 0.11093, Mean 0.00559, Std. 0.01552
IMPORTANT: THE
ORIGINAL VALUES WILL NOT Normal Score: Min. -2.7086, Max. 2.70860, Mean 0, Std. 0.99906
BE overwritten in the project.
A transformation table will
be created in the
background, like in the
example shown, and will be
used for transformation and
back-transformation.

342 • Petrophysical Modeling Property Modeling

Applying Trend Transformation
Exercise Steps
1. Continue with the same process from the previous exercise:
Exercise Model grid, Data analysis> PHI [U]> Zone A>
Facies_Adaptive_Model as the property to analyze.
2. Select a Facies type, for example, Channel; view the changes
in the histogram when a transformation is applied; by changing
from Input to Output distribution.

3. Add 1D trend before Normal score by using the Insert

transformation above the selected transformation button When applying a
trend transformation, click
to see if there is an obvious directional trend. Switch the Refresh histogram button
to update the result. If no
between upscaled cells and raw logs to perform the trend exists, remove the
analysis. trend from the
transformation list.

Property Modeling Petrophysical Modeling • 343

 Next, analyze the Crossplot from the 1D trend transformation. The
crossplot shows the correlation between Porosity (PHI) and Depth
(Z-values). You can see small trends in certain parts of the reservoir
caused by the effect of the logs coarsening/fining upwards. This effect
can be called local varying trend which is not possible to model.
4. After your own analysis, remove the 1D trend from the
transformation list by selecting the 1D trend transformation
and clicking on the Remove the selected transformation
button . Click Apply to save the changes.
5. Go to the Zone B, select Lobe facies code and make sure that
the Use logs button is selected. Visualize the histogram
input distribution and see that for the Lobe facies, the
distribution is bimodal.
344 • Petrophysical Modeling Property Modeling
 6. Add a 1D trend transformation. Pay special attention to model
the vertical trend for the Lobe facies, as shown in the next
figure. Note that the trend model (blue line) starts flat and then
changes to fit the points distribution. This is to avoid negative
values on the simulation result (remember that the line will be
extrapolated until it intersects the X-axis).
7. Another option to avoid negative values on the simulation In the
result is to add an Output truncation transformation and Transformation tab, the
Activate/Deactivate
specify Minimum value as zero for all the facies in both zones selected transformation
(A and B).
button can be used to
8. Select 1D trend transformation and change to the Histogram visualize the difference in
output distribution to see the transformation effect on the the histogram output without
data. Note that the histogram shows the distribution range deleting the transformation
entirely.
after trend removal, but before Normal score.

Property Modeling Petrophysical Modeling • 345

 9. Select the Normal Score transformation and click Refresh
histogram. The histogram will follow a standard normal
Normal score
transformation should distribution. Note that after trend removal, the Variogram can
always be the last be easier to interpret (see the two next figures for ZoneB and
transformation on the list.
Trends in the data will
facies Lobe showing the variogram with and without trend
remain after normal score removal).
has been applied unless they 10. Follow the same procedure to transform PHI [U] in all the
have been removed. Hence,
it is important to look for
facies and both zones to be standard normally distributed.
trends and remove them if 11. Click Apply to save the changes.
they are significant.

Figure: Variogram of the facies Lobe without 1D trend removal. In the

sample variogram, a cyclicity behavior with a trend can be seen.

Figure: Variogram of facies Lobe after 1D trend removal. In the sample

variogram can be seen the cyclicity but without trend effect.

346 • Petrophysical Modeling Property Modeling

Variograms
Variogram analysis is used to determine the spatial variation within a
property. The process includes experimental variogram generation and
variogram modeling. It is an interpretative process and you should also
take into account all available geological knowledge of the area.
Usually, vertical variogram models have plenty of data (assuming
vertical wells) and can easily be estimated. Horizontal variograms
normally have very little data and are usually implied from geology.

Vertical and directional variogram analysis for porosity

In this exercise, we will focus on the vertical variogram analysis for
porosity in each facies. Since permeability is closely correlated to
porosity, we will use the same variograms for permeability as porosity.
Exercise Steps
1. Continue from the previous exercise: Exercise Model grid,
Data analysis> PHI [U]> Zone A> Facies_Adaptive_
Model as the property to analyze. Change to the Variogram
tab under the Data analysis dialog.
2. Select the Vertical direction. It is recommended to start with
the vertical direction as it is easier to estimate, and it normally
also gives you the Nugget (same for all directions).
3. Specify the variogram settings according to the next figure. Use
Zone A and the Facies_Adaptive_Model as facies model
and Channel as Facies type. Define the Vertical range
(around 12) and the Nugget (around 0.25).
4. Change to the other Facies types and try to define the
variogram range and nugget also for these Facies types. Do
the same for Zone B.

Property Modeling Petrophysical Modeling • 347

 5. Change to Major direction. Define this direction to be 134
degrees (the Minor direction will be set perpendicular to the
Major direction).
6. Adjust other Search parameters, the resulting variogram will
automatically be updated. The recommended value for lateral
lag distance is based on distance between the wells.
7. Fit the Variogram curve (blue line) to the Sample
variogram (points). The Variogram ranges and Nugget will be
automatically updated. Note that the Nugget should be
derived from the Vertical variogram.

348 • Petrophysical Modeling Property Modeling

 8. Since the Horizontal variogram in this case cannot be
calculated from the available well data, it needs to be
estimated from the geological knowledge. Use the numbers
given in the table below for the Horizontal variogram model.
However, try to create your own Variogram model for the
Vertical direction, the values can be different for each Facies in
each Zone.

Facies Azimuth Major range Minor range Nugget Vertical range

Shale 134 3500 1500 0.1 7
Channel 134 3500 1500 0.25 12
Levee 134 3500 1500 0.2 12
Lobe 134 3500 1500 0.175 3

9. Click Apply to save the changes.

Data analysis process for Permeability using a Porosity

trend
In this exercise, you will apply a complete Data analysis workflow
(transformations and directional variograms modeling) for the
Permeability property. The exercise shows you a way to use the trend
between the logarithm of permeability and the porosity, removing it
from the log transformed permeability input, to later do the simulation
on the residuals. After simulation, the data will be back-transformed
and the trends added back. It is recommended to use the Data
analysis process for the transformation because you have visual
control over the process.
Exercise Steps
1. Open the Data analysis process and select the property Perm
[U]. Notice that ZoneA is already selected. Click the Unlock
button to open the process.
2. Visualize the histogram input for the upscaled cells and
raw logs . Realize the difference in the Permeability range
between the upscaled cells (0.01-607.38) and the raw logs
(0.01-1300). It can be a consequence of the Scale up well log
process. The actual upscaled value may differ from the above
depending on the Upscaling method.
3. Insert an Output truncation and specify the Minimum value
Property Modeling Petrophysical Modeling • 349
 to be 0 and Maximum value to be 1300. The effect of this
transformation can not be seen on the histogram output in
Data analysis process. The property will obtain that range
once the modeling process is performed.
4. Add a Logarithmic transformation to remove the logarithmic
character on the property and to get it in normal distribution.
5. Next, insert a 3D trend transformation to check if a trend
exists. Select the PHI_Data_analysis [U] property and
analyze the crossplot to model the trend as shown in the next
figure. Make sure to Scale with a general function to be
able to edit the blue curve (applied trend).

350 • Petrophysical Modeling Property Modeling

 6. Select the Normal Score transformation and click Refresh
histogram. The histogram Final will follow a standard normal
distribution. The input data must
first be log transformed to
7. Open the Variogram tab and try to model the Horizontal achieve a linear relationship
variogram. Otherwise, use a range of 3500 for the Major between porosity and
permeability! Use a 3D
and 1500 for the Minor. Define the Major direction to be 134
trend, using ‘PHI_Data_
degrees. Model the Vertical variogram. analysis’.
8. Click Apply to save the changes.
9. Make the same analysis for Zone B. The data analysis of
Perm (U) property will be used in the exercises for next
modules.

Property Modeling Petrophysical Modeling • 351

 Summary
This module covered the general concepts in Petrophysical modeling,
including Data analysis to prepare and QC data inputs. You learned how
to use different trends and how to identify the spatial variation for each
petrophysical parameter. Also various Transformations and Variogram
modeling was covered. As in Facies modeling, trends and secondary
information were used as additional input to guide the distribution in
3D.

352 • Petrophysical Modeling Property Modeling

Module 14 - Gaussian
Simulation in Petrophysical
modeling

The key issues for stochastic Petrophysical modeling using the

Sequential Gaussian Simulation (SGS) and Gaussian Random Function
Simulation algorithms will be covered in this module along with
property distribution for different facies as a conditional parameter.
Also covered is how to use correlated secondary data to identify the
anisotropy and spatial variation for petrophysical parameters.

Prerequisites
To successfully complete this module, the user must have knowledge of
the following:
• Familiarity with Geostatistical Fundamentals
• Familiarity with Petrophysics Fundamentals
• Familiarity with Reservoir Modeling
• Petrel Introduction

Learning Objectives
The purpose of this topic is learning how to perform the following
procedures within the related tools in Petrel:
• Run the Sequential Gaussian Simulation and Gaussian Random
Function Simulation methods to populate the model within
petrophysical properties.
• Use the Petrophysical modeling process to create different
models of Porosity.
• Use different output properties to correlate for possible
Co-kriging/Simulation.
• Use different methods of quality checking both the input data
and the resulting model.

Property Modeling Gaussian Simulation in Petrophysical Modeling • 353

 Lesson

Gaussian Simulation
Algorithms in Petrel

Sequential Gaussian Simulation

It is a widely used stochastic GSLIB method based on Kriging. It
can honor input data distributions, variograms and trends.

Gaussian Random Function Simulation

Faster than SGS and has better reproduction of input statistics
It is not sequential which allows the algorithm to be parallelized
using a decomposition:

Conditional simulation = Kriging +Unconditional Simulation

The unconditional simulation is a Fast Fourier Transform

algorithm which gives good variogram reproduction.

Two available Gaussian Simulation algorithms

In stochastic methods, the result is dependant on a random seed
number, and multiple representations are recommended to gain an
understanding of the uncertainty. The principal goal of stochastic
simulations is to also capture extreme values and produce “many”
different realizations to show the “true” variability of the existing data
set.
1. Sequential Gaussian Simulation (SGS) is a stochastic
(pixel-based) modeling algorithm that honors well data, input
distributions, variograms and trends. The variograms and
distributions are used to create local variations away from
input data.
2. Gaussian Random Function Simulation (GRFS) was
introduced in Petrel 2009.1. It is faster than SGS, has better
reproduction of input statistics through parallelization, fast
Collocated co-kriging and additional expert settings. Unlike
SGS, this algorithm is not sequential.
The algorithm itself works with a well-known decomposition,
using the new Kriging algorithm as a base. Unconditional
354 • Gaussian Simulation in Petrophysical Modeling  Property Modeling
 simulations can be very fast with the correct algorithm; the one
used here is a Fast Fourier Transform based algorithm which is
fast and parallelizable.
The interface for the new algorithm is similar to that of SGS except for
the Co-kriging and Expert tabs.
Common for both algorithms
For both algorithms, the Simple and Ordinary Kriging types are available
in the Petrophysical modeling process. For all of the versions of
Gaussian Simulation, the use of an external trend is available in the
Make/edit surface and Petrophysical modeling processes.
Note: The input data do not need to be normal (Gaussian) distributed,
because the data can be transformed. Back-transformation ensures that
the simulated values are in real units.
For more information about these methods, see the Petrophysical
modeling module (Sequential Gaussian Simulation and Gaussian
Random Function Simulation algorithms).

Gaussian Simulation
Data Transformation Input distribution
(well logs)

SGS and GRFS require:

1. Stationarity (no spatial trend; not

depending on location) Normal Score transformation
µ=0,s=1
2. Standard normal distribution (mean =
0, std dev = 1)

Simulation result will be automatically Back- transformation

back-transformed. Spatial trend and Output

original data distribution will be honored. distribution (3D

property)

Property Modeling Gaussian Simulation in Petrophysical Modeling • 355

 Normal score transformation
The concept of stationarity implies that the variogram model is valid no
matter what spatial location is being estimated or simulated. For
example, spatial statistics does not depend on location. When a trend
is present, the concept of stationarity is not really valid.
The transformations are applied in the order in which they are selected.
The simulation result will be automatically back-transformed in exactly
the reverse order. By doing so, the spatial trend observed in the input
will be added back and preserved. The original distribution will also be
honored. Output truncation is special in that it only applies to the final
property values after all inverse transformations have been applied.

Gaussian Simulation
Zone settings – Property & Variogram

Property, Zone and Facies model selection:

Make sure to pick up the correct upscaled
Property ([U] as suffix), the Zone and Facies
model.

Method selection:
Select SGS or GRFS as method for the zone.

Variogram :
Specify Range, Orientation, Nugget and Type
…or get a variogram from Data analysis

Petrophysical modeling – Zone settings tab

All of the settings that are specific for each zone are entered under the
Zone settings tab. This include the method/algorithm to be used for
each zone and the settings for the method that has been chosen.
Procedure:
1. Select the Property and Zone to model
2. Select the Method to be used
3. Specify the Settings for the chosen algorithm for the selected
Zone (including Variograms)

356 • Gaussian Simulation in Petrophysical Modeling  Property Modeling

 Gaussian Simulation
Zone settings – Distribution

Select Standard or Bivariate

…or get from Data analysis

If the property requires it, select

Is logarithmic

Used if secondary variable

with correlation is available

Used if upscaled logs

Used if no/few upscaled logs Used if upscaled logs

From Function
From distribution function window - Crossplot

Gaussian Simulation
Common settings - Realizations

Define Common settings for all zones: Realizations:

• Use filter Can be used to track changes in uncertainty
• Ensure that all cells get a value LOCATIONS in the grid, but in Uncertainty
• Number of realizations analysis we do not need them (as we just use
the seed as a variable)

Property Modeling Gaussian Simulation in Petrophysical Modeling • 357

 Petrophysical modeling – Common settings tab
This is where we apply Common settings for all of the Zones:
Use filter – Select this option if you want to use a filter. This can be
the Property filter (for example, filtered on certain facies codes), the
Zone filter or the Segment filter. Note that if you select this option, all
of the active filters will be used!
Ensure that all cells get a value – Selected by default. This will
force a value in any cell regardless of limitation in the chosen
algorithm.
Number of realizations – Select this and type in the number of
realizations to run. This should only be done after the Zone parameters
have been set up and tested once as running many realizations can be
time-consuming. If you have run a set of realizations, the Overwrite
function will overwrite these.

Gaussian Simulation
Result QC
QC results in a histogram:
Go to the Settings > Histogram tab for
the Property to check the distribution

Filter histogram:
Many types of filters can be used:
• Zone
• Raw logs, Upscaled cells and 3D
grid
• 3D grid property filter

358 • Gaussian Simulation in Petrophysical Modeling  Property Modeling

 Gaussian Simulation
Common settings – Local model update

Local model update: 2. Drop the polygon boundary inside the

This functionality allows for the updating of Petrophysical modeling > Common tab
a Facies or Petrophysical model in an
area inside a polygon. Original PHI (U) model

PHI (U) model with

Local model update
applied by using
new well data

1. Define a Polygon by using the Utilities >

Make/edit polygons process.

Common settings tab - Local model update

This functionality is available for Facies and Petrophysical modeling
processes.
Local model update - Is very useful for models that have been
validated and/or history matched, but where new well information
should update the model only locally (keeping the rest of the model as
is).
Only cells inside the polygon wall will be modeled when the process is
run, while keeping continuity with the neighboring (unchanged) cells
outside the polygon. The local model update is not run for zones where
object modeling is performed.
The polygon is arbitrary and can optionally consist of several
polygons grouped into one polygon object.
How to perform a local model update:
1. A Property model already exists and a new well has been drilled.
The new data needs to be included in the model.
Property Modeling Gaussian Simulation in Petrophysical Modeling • 359
 2. Run the Scale up well logs process to upscale the new log data
to the existing Property model. Select Edit existing and choose
the Property model to update. Also, select the Replace and add
new and Leave all other cells unchanged options. Make sure
that only the new well is selected in the Wells list.
3. Open the Property modeling dialog (Facies or Petrophysical). In
the Common tab, under Local model update section, use the
blue arrow to drop in the polygon from the Input pane. Set up the
lateral thickness in number of cells.
4. Set up all the modeling parameters and click OK to run the
process.

Gaussian Simulation
Petrophysical modeling – Using Secondary Data

Porosity is normally modeled prior to permeability:

- Porosity calculation is more reliable than permeability
- Porosity is better correlated to seismic attributes

Seismic attribute can be used as a secondary variable

A facies model can be used as input:

- The porosity model can be smoothed near facies body boundaries
to avoid sharp contrasts in the porosity values

Petrophysical modeling using secondary data

A secondary variable can be used to help distribute the petrophysical
property between well locations.
A standard Co-kriging can be used with the variogram derived both from
the primary property and from the dense secondary attribute.
Petrel uses Collocated Co-kriging. When using variograms from the
Data analysis process, the variogram from the primary data source is
used but no variogram from the secondary data is needed, as the other
control parameter is the correlation coefficient between the two
variables.

360 • Gaussian Simulation in Petrophysical Modeling  Property Modeling

 Gaussian Simulation
Correlation with Secondary data

Correlate two variables (primary to model and secondary as possible guide)

The number of data pairs is significant when defining a correlation between
two variables.
If correlation exist, Co-simulation with secondary variable is possible
A good correlation ensures a consistent model

Positive correlation Negative correlation No correlation

Correlation
Co-simulation with secondary variables usually assumes that the
second variable is available in a collocated form (that is, available at
the same cell location or positions as the cell which is about to be
simulated). The main aspect is that the primary data have few sample
points (for example, only a few upscaled well logs) while the secondary
data is present in all grid cells (like a resampled seismic attribute).
Crossplot analysis and computation of the correlation coefficient can
help identify useful secondary attributes or rule those out that show
little or no correlation. The magnitude of the correlation coefficient is
the thing to look for. A strong negative correlation is good as the
relationship does not have to show a positive correlation to be useful.
Crossplot in a Function window
In Petrel, a Crossplot can be displayed in a Function window. By
The actual
making a Linear Regression, the Correlation Coefficient can be correlation coefficient value
calculated. The new function will be stored in the Input pane (go to the is only displayed in the
Settings to see the parameters used such as the linear function). process window when
generating the function.

Property Modeling Gaussian Simulation in Petrophysical Modeling • 361

 Gaussian Simulation
Horizontal variograms from correlated attribute
Identify a correlated densely sampled attribute
Use a variogram map to examine anisotropy
Compute experimental variograms in the major and minor axes directions
Fit the model to the experimental variogram for the correlated attribute

SECONDARY DATA

Major Horizontal ranges

Variogram analysis
Horizontal Variogram analysis is used to determine the spatial variation
for a given property. In cases where sparse or very little data are
present, horizontal variograms frequently must be implied from geologic
knowledge or a densely sampled correlated attribute.

362 • Gaussian Simulation in Petrophysical Modeling  Property Modeling

 Gaussian Simulation
Variogram map & Sample variograms in Petrel
Horizontal variogram map – Decide direction of Anisotropy

Sample variogram – Decide Major and Minor Range

Major Minor
Variogram model

Simple variogram generation from Secondary property

Due to the number of samples, conventional variogram analysis using
the Data analysis process may be cumbersome. Variograms can also
be computed by accessing the settings for a given property. The
Variogram tab provides access to variograms for selected directions
along with variogram maps. This offers a more “batch like” approach to
computing variograms. For dense data, such as seismic attributes, it
may be impractical to try to use the Data analysis process. Any
change in the variogram parameters triggers a new computation. Even
with few grid layers, this could be a very time consuming analysis. The
Variogram tab can be used in conjunction with property filters to
optimize the run time and avoid CPU intensive interactive analysis.

Property Modeling Gaussian Simulation in Petrophysical Modeling • 363

 Horizontal variogram map
A recommended procedure is to compute a variogram map. This map
can then be displayed and examined to determine if sufficient
anisotropy is present. If present, the direction can be measured directly
on the map display using the measuring tool (located in the function
bar).
The Variogram tab can be used in conjunction with property filters to
optimize the run time and avoid CPU intensive variogram generations.
A variogram map is a way to present variograms that have been
computed in several different directions. This is done in an automated
fashion. The center of such a variogram map represents 0.0 lag
distance. Out from this center, the lag distances will show an increase
in several directions. Such maps are usually displayed as a surface in
Petrel. The grid geometry of such surfaces is in +/- lag space and is not
located in the project coordinate area. For this reason, they have to be
displayed in a Map window.
Steps:
1. Display the secondary property, use Index filer to filter.
2. In the property Settings>Variogram tab, select Horizontal
variogram map and toggle on use property filter.
3. Set up the parameters and press Run
4. Result is placed under 3D grid > Variograms folder
5. Display in a Map window

Sample Variograms
Once the major direction has been determined from the variogram map,
it is easy to produce sample variograms in the two directions (major
direction and minor perpendicular direction). The calculated variogram
model can be displayed in a Function window. The major and minor
ranges can be found in this Function window as well. This approach
avoids time-consuming variogram computation on large property grids
and can be used to uncover potential anisotropy, along with variogram
range in the major and minor direction.
When modeling, transfer the direction along with the major and minor
range to the respective primary property in the Data analysis dialog and
model the vertical variogram to obtain the nugget, sill and vertical
range values.
364 • Gaussian Simulation in Petrophysical Modeling  Property Modeling
Steps:
1. In the property Settings>Variogram tab, select Sample
variogram in direction (major) given by varigram map.
2. Set up the parameters and press Run
3. Placed under 3D grid > Variograms folder
4. Display in a Function window
5. Repeat the same procedure for minor direction (90 degree
difference) – make sure not to ‘Overwrite last’
When saved, this variogram will contain variogram parameters that
reflect the upscaled log data in the vertical direction, but have
horizontal parameters derived from the correlated secondary data via
the variogram map and sample variograms.

Gaussian Simulation in Petrophysical modeling–

Exercises
Usually, porosity is modeled prior to permeability because porosity
calculations are more reliable than permeability, and is better
correlated to seismic attributes.
A facies model should be used as input. In addition, you can use a
seismic attribute as a secondary variable.
The porosity model can be smoothed near facies body boundaries to
avoid sharp contrasts in the porosity values.
The 3D porosity property can later be used for conditioning when
modeling permeability in order to keep a consistent correlation.
In this section there is also an optional exercise showing you how to
model a net to gross property.
For your convenience, an extra project called Property Modeling
2010_M14.pet has been created with all necessary data in case you
have not completed any of the previous modules in the book. If you are
satisfied with your project, continue to work on it.

Property Modeling Gaussian Simulation in Petrophysical Modeling • 365

 Porosity modeling based on a Facies model using
Sequential Gaussian simulation method
Exercise Steps
1. Make a copy of the existing property ‘PHI (U)’ used during the
Data analysis process and rename it as ‘PHI_SGS’.
2. Make sure the Exercise Model grid is selected, open the
Petrophysical modeling process and select ‘PHI_SGS’.
3. Unlock the process and select the facies model ‘Facies_
Adaptive_Model’ using Sequential Gaussian simulation
as the method.
4. Select the Use the variograms made in data analysis
button. Notice that the Variogram parameters from the
Data analysis can be seen in the boxes, but these can be only
edited in the Data analysis process.
5. Select the Use the transformations made in data analysis
button.

6. Copy the same settings to all the other facies codes in Zone A

and Zone B by using the copy and paste buttons.

7. Under the Common settings, specify the Number of
realizations to be 3. Run the model by pressing OK. The
output will set a suffix [1] to the main property, and [2] and [3]
on the new realizations.
366 • Gaussian Simulation in Petrophysical Modeling  Property Modeling
 8. Use the I, J and K filters to visualize the model results.
9. Open the Settings for the Properties folder and go to the
Operations tab.
10. Select the 3 realizations of PHI and calculate a mean
porosity model by selecting Arithmetic mean as method.
Click on the button to create the new average
property.

11. Check the result: Visually compare the input Facies property
and the resulting mean Porosity property. Also, compare the
histogram between the upscaled cells, well logs and 3D
property for each facies. Set the property filter for each of the
facies as shown in the next figure:

Property Modeling Gaussian Simulation in Petrophysical Modeling • 367

 12. Look at the Histogram and Statistics tabs under the
Settings of the different Porosity models and apply the
property Filter by pushing the filter button in both tabs:

368 • Gaussian Simulation in Petrophysical Modeling  Property Modeling

 13. Remove all of the filters and make a copy of the mean porosity
model. Open the Settings for the copied model.
14. Go to the More tab and select ‘Facies_Adaptive_Model’ as
the facies model from the drop-down list.

15. Select Levee as Facies type 1, Shale as Facies type 2 and

Smooth only the cells where the facies is type 2.

Property Modeling Gaussian Simulation in Petrophysical Modeling • 369

 16. Click on the button. Visualize the two
different mean porosity models in a 3D window to see the
difference between them.

Optional Exercise: Net-to-Gross modeling using Gaussian

random function simulation method
Follow the steps below to generate the NtG model:
In the Exercise Exercise Steps
Model grid, a NtG_cont_
synt(1)(U) and NtG_ 1. Open the Petrophysical modeling process and select
Arithmetic mean property NtG_cont_synt property.
already exist. The NtG_cont_
synt(1)(U) property will be 2. Unlock the process and select the Gaussian random
used in the Volume function simulation algorithm as the method to use.
calculation process, and it
is therefore important to
3. In the Variogram tab, define the parameters:
populate it in 3D. • Variogram type: Spherical
• Nugget: 0.1
• Ranges Major (3500), Minor (1500) and Vertical
(10)
• Azimuth (134)

4. Copy the same settings to Zone B by using the copy and

paste buttons.
5. Specify the Number of realizations to be 5 and run the
model by clicking OK.
6. Open the Settings for the Properties folder and go to the
Operations tab.
7. Select the 5 realizations of NtG and calculate a mean property
model by selecting Arithmetic mean as method. Click on the
button to create the new average property.
8. Use the I, J and K filters to visualize the model results.

370 • Gaussian Simulation in Petrophysical Modeling  Property Modeling

Also, NtG modeling can be performed conditioning to a Facies model
(‘Facies_Adaptive_Model’) using the variogram parameters defined in
the Data analysis process. This way a Channel belt can be split into Net
and No-net intervals as most channels are not “full of sand”.

Optional Exercise: Correlation with secondary data

The correlation between two parameters can be captured when
Petrophysical modeling is performed. Correlation analysis can help you
to understand the relationship between two parameters and define the
correlation coefficient.
Since the correlation coefficient can be automatically estimated during
Petrophysical modeling for each facies, this exercise aims at providing
you with an understanding of the correlation between porosity and
permeability.
Exercise Steps
1. Open a Function window.
2. In the Properties folder, select ‘PHI_SGS’ (e.g. PHI_SGS(1))
on X-axis and ‘Perm’ on Y-axis, and select the third
parameter ‘Facies_Adaptive_Model’ for color.
3. Use log transform for Y-axis (Perm) and upscaled cells
.
4. Click on the Make Linear function from crossplot
Property Modeling Gaussian Simulation in Petrophysical Modeling • 371
 button to get the correlation coefficient and add a best fit
linear function through the data points.

5. Open a new Function window.

6. Also, check the correlation between AI and PHI by displaying
‘PHI_SGS’ on X-axis and ‘Al’ on Y-axis. A negative
correlation can be observed.
7. Next, filter part of the data displayed in the crossplot by
clicking the Select using 1D range on X axis button
(this will filter for porosity).
8. Open a new 3D window and display different properties
(‘PHI_SGS’, ‘Perm’, ‘Facies_Adaptive_Model’ etc) from
Models pane>Exercise Model>Property folder and apply
the generic filters by selecting them from the Filter folder in
the Input pane.
9. To see the effect of applying the generic filters, first tile the 3D
window and the Function window vertically or horizontally.
Then, apply the filters from the Function window to the 3D
window, filtering the 3D models.

372 • Gaussian Simulation in Petrophysical Modeling  Property Modeling

Variogram analysis using correlated attributes
Variogram maps are useful in determining anisotropy, and if data is
scarce, determining in which direction there is enough data to make a
stable sample variogram.
In the following exercises, we will examine the use of secondary data
sources for horizontal variogram analysis. Since we know that there is a
correlation between Porosity (PHI) and Acoustic Impedance (AI), the 3D
property AI can be used for getting variogram information (major and
minor directions) for later use in porosity modeling.
First, we must create a variogram map to get information about the
orientation of the variogram. The second step is to use this information
when establishing a sample variogram. The final step is to fit a
variogram model to the sample variogram.
Creating a variogram map from an Acoustic Impedance
property
The AI property exists in the entire 3D grid and shows that the data are
not completely random. Rather, they are distributed according to the
geological features they represent. A correlation between the AI and
PHI logs have been observed, hence the AI property may be used to
obtain information about the variogram.

Property Modeling Gaussian Simulation in Petrophysical Modeling • 373

 Exercise Steps
1. For the Exercise Model, show the AI property.
2. Use a property filter on the AI property during the calculation:
This will help speed up the computation without decimating
the AI data too much.
3. We recommend using the Index filtering parameters from
Properties folder>Settings>Filter tab, to limit the computing
time. Apply the settings as shown in the next figure.
Deselect the And
option (If this is selected,
only the intersections
between the selected I and J
cells will be shown as
pillars).

4. Display the AI property in a 3D window to verify the effect of

the Index filter.

5. Go to Settings > Variogram tab for the AI property. Select

the Horizontal variogram map option. On the XY range tab,
specify the parameters for Number of lags x and y to 10,
374 • Gaussian Simulation in Petrophysical Modeling  Property Modeling
 Search distance in X- and Y-direction to 5000. In addition,
restrict the Z range to None.
Comment
The Z range when making Horizontal variogram maps means that the
pair of points which are not in the same layer can be consider for
the variance calculation. It is similar to the vertical Thickness in
the Data analysis process dialog, for Major and Minor directions
variograms. If the None option is chosen, only pairs of points belonging
to the same layer have the variance computed.

6. Make the variogram map by clicking the Run button. The new
Var. Map from AI will be stored in Exercise Model > Once completed, a
Variograms folder. variogram map object will
appear under the Variograms
7. Display the variogram map computed from the AI property in folder located in the Models
a Map window. Adjust the contour interval to reveal the pane.
anisotropy.

Property Modeling Gaussian Simulation in Petrophysical Modeling • 375

 8. Using the Measure Distance button, place a line from
the center on the map following the major trend axis and
record the Heading or azimuth reported in status bar beneath
the Map window (azimuth should be approximately -45
degrees which equals a heading of 135 degrees).

Creating sample variogram from an Acoustic Impedance

property
A sample variogram is a plot of separation distance against semi-
variance for the data observed. The sample variogram must be
established before a mathematical model (the variogram) can be fitted
to those data.
The orientation obtained in the previous exercise will now be used as
input for establishing the sample variogram.
Exercise Steps
1. Return to the Variogram tab and select the Sample
variogram option. Select the option Horizontally.
376 • Gaussian Simulation in Petrophysical Modeling  Property Modeling
 2. On the Orientation tab, deselect the Isotropic variogram
check box and adjust the Orientation to match the value
recorded in the previous step (-45). This will allow you to
compute an experimental variogram for the major axis
direction. Adjust the angular Tolerance to approximately 30
degrees. This will ensure that sufficient directional preference
is used to collect data pairs during the variogram computation

3. On the XY range tab, increase the Number of lags to 20 and

Horizontal search radius to 5000
4. On the Z range tab, use None as the vertical search distance.
5. Then, Run the variogram computation. When completed, a
sample variogram object will appear in the Variograms folder
in the Models pane>Exercise Model.
6. Repeat this procedure after adjusting the orientation 90
degrees (from –45 to 45 degrees). This will allow you to
compute the sample variogram for the minor axis direction as
well. Make sure not to select the Overwrite last check box
since you are generating a new sample variogram.
7. Open a Function window and display the experimental
variograms for the major and minor directions.

Property Modeling Gaussian Simulation in Petrophysical Modeling • 377

 Creating a variogram model based on the sample
variogram
A variogram model is a mathematical representation of the
experimental variogram. It is described by the variogram settings such
as range, sill, nugget and orientation. The orientation for the Acoustic
Impedance (AI) property was found from the variogram map. The ranges
will be established by inspecting the sample variograms for the major
and minor directions.
Exercise Steps
1. Continue from the previous exercise (display ONLY the sample
variogram for the major direction in a Function window).
2. Make a new variogram model under the Variograms folder by
clicking the Make variogram for Sample Variogram
button in the function bar. It is stored in the Variograms
folder.
3. Rename this new variogram object to “Var. from AI”.
4. On the Style tab, make sure the check boxes for showing both
378 • Gaussian Simulation in Petrophysical Modeling  Property Modeling
 the min and max direction are selected as shown in the next
figure. Display this variogram model in the Function window
along with the sample variograms previously computed for the
major and minor directions.

5. On the Settings tab, adjust the Orientation angle as shown in

the next figure.

6. Display the variogram model in the Function window along

with the experimental variograms for the major and minor
axes.
Property Modeling Gaussian Simulation in Petrophysical Modeling • 379
 7. At this point, adjust the point representing the (range, sill)
value for both the major and minor direction to better fit the
The Variogram model sample variogram data. This can be done graphically by first
parameters calculated from a
AI will be used in the selecting either Select and Edit line or Select and
exercises for next modules to
Edit/Add Points buttons in the function bar and then
model a Porosity property.
editing the points in the Function window.

Comment
At this stage, you do not have to be overly concerned about the sill and
nugget values. When the vertical variogram is analyzed using the
Porosity data, we will determine the nugget and sill values to use in our
final variogram model. The current goal is to establish ranges in the
major and minor directions from the correlated secondary data source.
You should also realize that the secondary data has not been subjected
to any transformation and may exhibit some degree of trend. In the
figure under point 6, you will notice the drift in the semi-variance values
with the larger lag distances. This drift suggests there is a trend
present. For this reason, the major and minor ranges have been chosen
based on the shortened lag values.
The vertical aspect of the porosity variogram can be derived directly
from the upscaled log curve data values as you have done in previous
380 • Gaussian Simulation in Petrophysical Modeling  Property Modeling
exercises. For the horizontal direction, simply assign horizontal ranges
and orientation that were recorded during the AI variogram modeling
steps above. These parameter values modeled for AI will be used for
horizontal variograms for porosity.

Local model update of a Porosity model

In the model area, a new well has been drilled (Test 2) and its log data
needs to be included on the existing Porosity model.
The Local model update will be done by using the variogram parameters
calculated from the secondary data (Acoustic impedance property).
1. Make sure the Exercise Model grid is selected. Go to the
Properties folder and make a copy of the existing property
‘Porosity (U)’ and rename it to ‘Porosity_LMU (U)’.
2. Open two new 3D windows and tile them vertically. Display
both models Porosity (U) and Porosity_LMU (U).
3. In each window, also display the Test2 well from
Wells>Other Wells folder and the Test 2 polygon stored into
the Project Polygons folder in the Input pane.
4. Go to Properties folder>Filter tab, in the option Upscaled
select Special 1, to visualize a 1 cell filtered around the
upscaled cells.
5. Open the Scale up well logs process, select the option Edit
existing and the Porosity_LMU (U) model.
6. Choose the settings Replace and add new and Leave all
other cells unchanged to upscale the Test2 PHI log. In the
Wells list, only select the Test2 well. Click OK.

Property Modeling Gaussian Simulation in Petrophysical Modeling • 381

 7. Open the Petrophysical modeling process and select
‘Porosity_LMU (U)’. Click on the Common button. In the
Local model update section, select the Only overwrite
cells inside polygon check box. By using the blue arrow,
insert the Test 2 polygon.

382 • Gaussian Simulation in Petrophysical Modeling  Property Modeling

 8. Use variogram model (Var. from AI) parameters calculated
from the secondary data (AI) stored in Exercise Model>
Variograms folder. Open the variogram model
Settings>Settings tab to visualize the parameters and type
them into the Zone settings>Variograms tab for both Zones
(A and B).
9. Click OK to run the process. Visualize the result on the
displayed model by using the IJK property filter player.

Property Modeling Gaussian Simulation in Petrophysical Modeling • 383

 Summary
This module covered the key issues for stochastic petrophysical
modeling using the Sequential Gaussian Simulation (SGS) and Gaussian
Random Function Simulation (GRFS) algorithms. This includes property
distribution for different facies as conditional parameters, the use of
correlated secondary data to help distribute the primary data into the
model, and the use of variograms to include any anisotropy and spatial
variation for the petrophysical parameters. In addition, it was also
shown how the horizontal variogram parameters could be derived from
a correlated seismic attribute (acoustic impedance).

384 • Gaussian Simulation in Petrophysical Modeling  Property Modeling

Module 15 - Kriging in
Petrophysical Modeling

The key issues for petrophysical modeling using the Kriging algorithm
will be covered in this module along with property distribution for
different facies as a conditional parameter, trends and Co-kriging. The
use of Kriging algorithm is focused on water saturation modeling to
show the strength of the algorithm.

Prerequisites
To successfully complete this module, the user must have knowledge of
the following:
• Familiarity with Geostatistical Fundamentals
• Familiarity with Petrophysical Fundamentals
• Familiarity with Reservoir Modeling
• Petrel Introduction

Learning Objectives
The purpose of this topic is to give the participant an introduction to the
Kriging algorithm. At the completion of this module, you will be able to:
• Realize the advantages and limitations of using the Kriging
algorithm
• Run the Kriging algorithm to populate the model with
petrophysical properties
• Use the Kriging settings to create a water saturation model
using conditioning, trends and Co-kriging
• Quality check the output model using different methods

Property Modeling Kriging In Petrophysical Modeling • 385

 Lesson

Kriging in Petrophysical modeling

Kriging algorithm
Overview
Kriging is a deterministic estimation algorithm based
on fundamental statistical properties of the data
(Mean and Variance).
Input:
- Well data, facies model, trends, secondary variables
- Variogram model
Methods:
– Simple Kriging
– Ordinary Kriging
Output:
– One estimation model

Three Kriging algorithms are available in Petrel:

Kriging Interpolation: It searches data in X,Y,Z and populates data in
I,J,K. It only considers data within the variogram range. It may show
strange effects in areas with no data when trends have not been
removed correctly.
Kriging by Gslib: It works in I,J,K space, has Collocated co-kriging and
expert settings. But its performance is slower compared to ‘Kriging’.
Kriging: Introduced in Petrel 2008.1. It is fast and is able to krige much
larger numbers of data through a combination of multithreading and
smart neighborhood searching. In Petrel 2009 and 2010 versions, it has
seen considerable improvements through parallelization, fast
Collocated co-kriging and expert settings. It works in XYZ and IJK
space.
The ‘Kriging’ method user interface is similar to the ‘Gaussian random
function simulation’ except for the Distribution and Expert tabs.

386 • Kriging In Petrophysical Modeling Property Modeling

 Kriging in Petrophysical modeling
Variogram & Distribution set up

Select Kriging as method to use

Variogram tab:
A. Specify Range, Orientation, Nugget
and Type
B. …or get a variogram from Data
analysis

Distribution tab:
A. By default the distribution taken from
the upscaled logs
B. …or use from Data analysis

Kriging in Petrophysical modeling

Co-Kriging using Secondary data
Primary
The Kriging method has been improved
and a Co-kriging option is added. It
Can use a Secondary variable:
• Property
• Horizontal surface
• Vertical function

The linear correlation coefficient can be Secondary

estimated. A fast on-the-fly correlation using
the interactive slider bar can be performed.

Property Modeling Kriging In Petrophysical Modeling • 387

 Co-kriging
The Kriging algorithm has a fast Collocated co-kriging option that is
implemented using a known decomposition which allows splitting into
Kriging (which is done once) and a simple Bayesian co-kriging update.
For any defined correlation between the primary and secondary
variable, a simple linear weighting is enough to give the exact value of
the Collocated co-kriging term. Due to the fact that it is a rapid process,
it has been implemented on a slide bar.
For detailed information about Collocated co-kriging function see
Module 16.

Kriging in Petrophysical modeling

Collocated Co-Kriging – Slider bar

How does the slide bar work?

First select the secondary property and click Estimate to get the linear correlation coefficient.
Click Apply to run the Kriging. Once it is run the correlation coefficient can be updated with the
result visualized on-the-fly using the slider bar.

Slider bar
Notice that the slider bar for fast Collocated co-kriging is activated after
the modeling process is performed. Then, the correlation coefficient can
be adjusted interactively by dragging the slider bar.

388 • Kriging In Petrophysical Modeling Property Modeling

 Kriging in Petrophysical modeling
Trends using Secondary data

Trends can be used as correlated

Secondary data as well:
• Property trend (3D)
• Horizontal trend surface (2D)
• Vertical trend function (1D)

The trends can be included into the

process by selecting them in the Trends
tab or using Data analysis process.

Kriging in Petrophysical modeling

Expert settings

Additional Expert settings

have been added to extend and improve the control over the Kriging output.

IJK interpolation

XYZ interpolation

Property Modeling Kriging In Petrophysical Modeling • 389

 Kriging Expert settings
Expert settings have been added to extend user control over the style of
Kriging:
• Can be performed with the options ‘IJK kriging (conform to
grid layering)’ or ‘XYZ kriging (conform to sea level)’.
• The Layer declustered search option ensures that when the
Kriging algorithm is searching for the nearest neighbors of a
cell which is to be kriged, it preferentially searches for
neighbors in the current layer and then progressively for
neighbors in nearby layers.
• The Use approximate search provides a search algorithm
which can often be substantially faster than the standard
search algorithm. A typical application is to use it for the time
consuming activity of creating many realizations for Kriging.

Kriging In Petrophysical Modeling - Exercises

Geoscientists and Reservoir Engineers use different methods to model
Water Saturation (Sw). In this exercise we demonstrate one option for
modeling SW using the different settings of the Kriging algorithm.
This exercise also covers the importance of the variogram model, as
well as use of secondary data like porosity model and property trend to
condition the Sw property modeling.
For your convenience, an extra project called Property Modeling
2010_M15.pet has been created with all necessary data in case you
have not completed any of the previous modules in the book. If you are
satisfied with your project, continue to work on it.
Modeling a SW property using Kriging method and
secondary information
Exercise Steps
1. Make sure the Exercise Model grid is selected and open the
If you have not Petrophysical modeling process.
created the property earlier, 2. Select the Existing property ‘SW_Archie (U)’.
copy the SW_Archie 3. Select Kriging as the method to use for Zone A.
property form the ‘Sw
Models’ folder in the 4. Go to the Variogram tab and define the parameters as shown
Properties folder under the in the figure below:
Complete Model and paste
it in the Exercise Model.
390 • Kriging In Petrophysical Modeling Property Modeling
 5. Run the model by clicking Apply. Display the result in a 3D
window.
6. In the Intersection folder>Right-click>Insert general
intersection, make it parallel to the I-direction and use the
Toggle visualization on plane button to display the
‘SW_Archie’ model. Also, display the model in the 3D
window and use the K-filter to visualize the result. To get a
better view, also use vertical exaggeration of Z to 10.

The Sw calculation
normally follows the sea
level rather than the layers.
Therefore, it can be a good
idea to use an algorithm
which handles XYZ modeling.
This functionality is found
under the Expert tab when
using Kriging algorithm.
7. For both zones, go to the Expert tab and select the XYZ
Property Modeling Kriging In Petrophysical Modeling • 391
 kriging (conform to the sea level) option and click OK to
rerun the model for both zones.

8. Use the General intersection and the I, J and K-filters to

check the result. Visualize to make sure the SW_Archie
distribution is following the “sea level”.

9. Before you use secondary data for modeling, the correlation

between both properties must be verified. If a correlation
exists, the secondary data can be used for Co-kriging. Open a
new Function window.
10. Display the ‘Porosity_LMU’ property on the X-axis, ‘SW_
Archie’ property on the Y-axis and ‘Facies_Adaptive_
model’ as Z-values. Make sure you view the Upscaled cells
only.
392 • Kriging In Petrophysical Modeling Property Modeling
 11. Make the linear function from crossplot to check the
correlation coefficient between both properties.

12. Go the Petrophysical modeling process and make sure that

‘Sw_Archie’ property is selected.
13. For both Zones A and B, go to the Co-kriging tab, check on
Secondary variable>Property and select the ‘Porosity_
LMU’ property. Click on the Estimate button to calculate the
Collocated co-kriging coefficient.

14. Click Apply to run the model.

Property Modeling Kriging In Petrophysical Modeling • 393
 15. Visualize the ‘SW_Archie’ property result, using General
intersection, Well section fence or I, J and K filters.

16. Return to the Co-kriging tab and vary the correlation

coefficient by moving the slider bar until the desired effect of
the Porosity property is visible on the ‘Sw_Archie’ property.

17. In the Petrophysical modeling process, click on the Show

data analysis dialog button. In the Data analysis,
unlock the process to access the settings dialog.
18. In the Transformations tab, insert a 3D trend transformation
and select the Above contact property.
394 • Kriging In Petrophysical Modeling Property Modeling
 19. Make sure you check Scale and Scale with a general
function, to have full control on your model trend (blue line).
20. Do this for both Zones A and B.

21. Make sure to add an Output truncation to get the desired

range for ‘SW_Archie’ after modeling (SW = between 0-1).
22. Click OK to save the changes in Data analysis.
23. In the Petrophysical modeling process, for Zone A and
Zone B, click on the Use the transformation made in data
analysis button.
24. In the Co-kriging tab, keep the Secondary variable >
Property > Porosity model and its Collocated co-kriging
coefficient. Click Apply to run the model.
25. Adjust the correlation coefficient by moving the slider bar if
needed and visualize the result.
Property Modeling Kriging In Petrophysical Modeling • 395
 The property ‘SW_Archie’ will be used in the Volume Calculation in
Module 17.

Optional: By deselecting Use the transformation made in data

analysis and Co-kriging, the Trends tab becomes available. Select
the check box Property trend and select the ‘Above contact’
property. Run the modeling process and check the result.

396 • Kriging In Petrophysical Modeling Property Modeling

Summary
In this module, you learned that the Kriging algorithm is a useful option
for modeling continuous properties such as water saturation (SW). It
was demonstrated that Collocated co-kriging is helpful for correlated
properties where one property is sensitive to the input parameters and
can be stabilized via the correlation coefficient to an already existing
property. It was also shown that trends can be used to guide the
population.

Property Modeling Kriging In Petrophysical Modeling • 397

Module 16 - Petrophysical
Modeling using Secondary
Data

The use of secondary information through Collocated co-kriging, Local

varying mean and Bivariate distribution in the Petrophysical
modeling process are covered in this module. The main focus will be
how to use different methods and to identify the spatial variation for
each petrophysical parameter and its correlation.

Prerequisites
To successfully complete this module, the user must have knowledge of
the following:
• Familiarity with Geostatistics Fundamentals
• Familiarity with Petrophysics Fundamentals
• Familiarity with Reservoir Modeling
• Basic/intermediate Petrel knowledge

Learning Objectives
The purpose of this topic is to give the participant an understanding of
the use of secondary information for Petrophysical modeling and to
become familiar with the related tools in Petrel.
At the completion of this training, you will be able to:
• Prepare and quality control secondary data using the Petrel
Data analysis process
• Use different output properties to correlate for possible
Co-kriging simulation
• Run Gaussian Simulation algorithms using secondary
information to distribute petrophysical properties, such as
Porosity and Permeability
• Quality check both the input data and the resulting model

Property Modeling Petrophysical Modeling using Secondary Data • 399

 Lesson

Petrophysical Modeling using Secondary Data

Gaussian Simulation and Kriging – Secondary data in modeling

Local Varying Mean: uses Simple

Kriging and treats secondary data as
location-dependant mean.

Collocated Co-Kriging: includes the

correlation coefficient of the secondary
data in the Kriging equation.

Bivariate Distribution: A secondary

variable is used to define bins, splitting
the distribution of the primary variable
into several intervals using separate
CDF’s for back-transformation.

400 • Petrophysical Modeling using Secondary Data Property Modeling

Petrophysical Modeling using Secondary Data
Local Varying Mean (LVM)
n n
z ( x0 ) = ∑ li z ( xi ) + [1 − ∑ li ]m( x 0)
i i
Z(xi): Data points (e.g. porosity)
m(x0 ): Secondary input such as a 2D map (porosity) or a property with a strongly
correlated positive value
The sum of the weights li may be less than one
The smaller the weights the bigger the influence of the local varying mean m(x0) on
the calculated value Z(x0)
Local varying mean gains influence with increasing distance of location x0 from
data points (decreasing weights i ) l
IMPORTANT Secondary input should be:
- Smooth and available for all locations x0
- Positively correlated to the primary data

Local Varying Mean (LVM)

In Gaussian Simulation algorithms, LVM uses secondary data as
local mean (m) values for the primary data. It uses Simple Kriging at
base, but the mean varies at each location rather than being constant.
With few primary data in one place, the weights (lambda) will be small,
so the kriged value will be close to the local mean.
It is required that the secondary data has the same units as the primary.
If different, the units should be converted to the same units as the
primary data.
Example: The local varying mean will generally be applied when the
primary and secondary data are the same. If you model porosity and a
(simple) porosity map already exists, with different areas of different
means, it is an example of how to use LVM.

Property Modeling Petrophysical Modeling using Secondary Data • 401

 Petrophysical Modeling using Secondary Data
LVM example - Permeability modeling using Porosity (small range)

Top: Permeability Map (Local Varying Mean)

Top right: Permeability vs Porosity xplot
Right: Histogram of Permeability

With decreasing variogram range, secondary

data (Porosity) gains control over the output

Units
Remember that it is a requirement that the secondary data has the
same units as the primary, or that the secondary data are strongly and
positively correlated to the primary data. The range referred to in the
slide relates to the horizontal variogram ranges; the smaller the range
of the primary data, the more the secondary data controls the modeling
result.
Transform using input distribution
Local Varying Mean There is an option to transform the secondary data into normal-score
will not work for negative
correlation.
space (if not used, the secondary variable will be transformed using a
normal distribution). A normal score transformation is used on both the
secondary property and the upscaled cells before calculating the
residual in normal score space. The residual will then be modeled and
the trend added before the back-transformation.

402 • Petrophysical Modeling using Secondary Data Property Modeling

Petrophysical Modeling using Secondary Data
Co-kriging – General background theory

Traditional Co-kriging equation:

Z C O K( x 0 ) = ∑ l i Z ( x i ) + ∑ µ j Y ( x j )
i j

Requires variograms of primary and secondary property, and a cross variogram

Consequently a larger equation system is constructed with more constraints

Collocated co-kriging: Z CCOK ( x0 ) = ∑ l Z ( x ) + μY(x )

i
i i 0

Possible solution in case of a more densly sampled secondary variable

Requires variogram only for the primary property, using a Correlation Coefficient for
the secondary property
A simpler equation system is constructed

What is Co-kriging and why do we use it?

The goal of Co-kriging is to predict an unknown property at a location
based on values of this (primary) variable combined with another
variable (secondary) which is present “everywhere” and cross-
correlated with the primary. It is a popular technique because it is good
at using data from, for example, a resampled seismic attribute (high
sampling) to predict porosity at the well position (low sampling)
Co-kriging
Regular co-kriging uses a generalized linear regression involving well
data (primary) and for example seismic data (secondary). As with kriging
it will minimize the error variance to determine the kriging weights. It
can be a cumbersome and tedious process as it must solve 3 equation
systems - and does not always add value to the simulation result.
Since it needs variograms for both the primary and secondary property,
and the cross variogram, it is a tedious calculation which can be
difficult to interpret and fit.

Property Modeling Petrophysical Modeling using Secondary Data • 403

 Collocated co-kriging
A simplified co-kriging method has been developed. The Markov model
performs a simplification, using only the primary variable variogram.
Then it uses a correlation coefficient of the already modeled secondary
variable. This method is simple and fast. Only the collocated co-kriging
option is available in Petrel.

Petrophysical Modeling using Secondary Data

Collocated co-kriging – Characteristics

A simplified equation system

 faster than traditional Co-kriging
No instability caused by highly redundant secondary variable
Control parameters: secondary variable, correlation
coefficient and variance reduction factor
Only the variogram of the primary variable must be modeled

Collocated co-kriging characteristics

Using the Collocated co-kriging method simplifies the Co-kriging
system, as we do not need to model cross-variograms and direct the
secondary variogram.
Basic assumption
• Densely sampled secondary variable
• Cross variogram proportional to variogram of primary variable
• Correlation
Control parameters
• Secondary variable with number of available data
• Correlation coefficient
• Variance reduction factor
Requirements
• Transformation of both variables to normal scores
• Calculation of correlation coefficient of normal scores
• Co-kriging with collocated secondary data
• Simulated value is obtained from a Gaussian distribution with
- Mean = Collocated co-kriging estimate
- Variance = Collocated co-kriging variance
404 • Petrophysical Modeling using Secondary Data Property Modeling
 • Variance reduction factor must be tuned to reproduce the
global variance

Petrophysical Modeling using Secondary Data

Collocated co-kriging – Control parameters
Number of data Correlation coefficient Variance Reduction Factor
Complete 3D grid CC = 0.8 VRF = 1.0

Every 3rd cell CC = 0.5 VRF = 0.5

Every 10th cell CC = 0.2 VRF = 0.2

Number of data
• Mostly porosity to support permeability because of good
correlation
• 3D porosity population first (easy to treat normal distributed
porosity; very often trend, better variogram etc.)
• The quality of the resulting primary property depends on the
number of data of the secondary data
Correlation coefficient
• Describes the relationship between both variables
• Linear coefficient between the normal scores
• Outlier sensitive
• Correlation Coefficient (CC) [-1 , +1]
• Depends on the number of data

Property Modeling Petrophysical Modeling using Secondary Data • 405

 Variance Reduction Factor (VRF)
• Kriging algorithm produces not only an estimation value but
also an estimation variance
• Not important for estimation purposes but for (co)simulation
• Determines the variability of the primary variable
• No way to determine, i.e., trial and error required
The Variance Reduction Factor is used to reduce the error variance
that is computed by the Co-kriging method. The Co-kriging algorithm
does not use the error variance, but the error variance defines the
spread of the distribution from which simulated values are drawn by
the Gaussian Simulation methods. Since the error variance is frequently
over-estimated, the result can be improved by multiplying it with the
Variance Reduction Factor.
For the Sequential Gaussian Simulation method, the variance of the
simulated primary variable is often systematically different from the
desired input variance. In the typical case where the secondary variable
is smoother than the primary (often the secondary variable is a smooth
seismic data set), the tendency is for the variance of the simulated data
to be higher than expected. Furthermore, the calculated correlation
between the simulated primary variable and the secondary variable is
not equal to the input correlation. This is a problem associated with the
sequential nature of SGS. In Petrel, the mechanism to overcome such a
bias is the variance reduction factor. This value can be selected by
trial and error to partly remove the bias.
For Gaussian Random Function Simulation method, this bias is
removed leading to a better recovery of input histograms and
correlation between primary and secondary variables. The fast
Collocated co-simulation is implemented in a manner similar to the fast
Collocated co-kriging.

406 • Petrophysical Modeling using Secondary Data Property Modeling

Petrophysical Modeling using Secondary Data
Permeability modeling – Using 3D Trend (Data analysis)

Data Analysis allows control of the transformations

Note: Use simulation with

Trend only if the secondary
input is smooth!

3D trend
The smooth trend property is specified and the deterministic function
generated (possible to edit).
Once the modeled function (blue curve) is ready, the trend is removed
and the residuals are ready to be modeled.
You should always quality check the values of the resulting model by
displaying them in a histogram. If extreme values exist, these can be
removed by doing an output truncation of the data. The reason for
scaling the data with a non-linear function (as in the slide above) is to
avoid extreme values in the resulting 3D property.

Property Modeling Petrophysical Modeling using Secondary Data • 407

 Petrophysical Modeling using Secondary Data
Bivariate Distribution – used in back-transformation

Bivariate distribution needs a secondary property covering the same area as

the area to be simulated and a crossplot of the two variables divided in classes
Distribution of primary property will depend on value of a secondary property. The
provided distribution decide how to back-transform the normal-scored data
initially produced by Gaussian Simulation algorithms

Porosity vs Log Permeability xplot

Bivariate distributions
When performing a Bivariate distribution, the user also supplies a
secondary property covering the same area as the area to be
simulated, and a crossplot of the two variables. In the result, the input
distribution will be honored as before, but the crossplot of the two
variables will be honored and the primary variable (that is modeled) will
follow the same general pattern as the secondary variable.
This is achieved during the back-transformation of the simulation
results. The forward transformation prior to modeling will be the same
as with the univariate transformation (Standard distributions and
Normal Score transformations).
By default (if no ranges are set), Petrel will split the crossplot into 10
bin intervals based on the secondary property, each with an equal
number of points. The user can decide the amount of bins in the
Function window. For each bin a separate distribution will be
calculated.
408 • Petrophysical Modeling using Secondary Data Property Modeling
The back-transformation will then be performed according to which bin
interval the secondary property in the appropriate cell falls into. There
is an option to use the upscaled logs or a crossplot. The crossplot is
more versatile as you can edit the cuts (deciding the bins) while this is
done automatically using upscaled logs and a specified cut number.

Petrophysical Modeling using Secondary Data

Bivariate Distribution - Permeability CDF as function of Porosity

CDF for low porosity CDF for medium porosity CDF for high porosity

If you split the input data into three different classes, it means that
three different Cumulative Distribution Functions (CDF) will be
used during the back-transformation of the simulation results. As the
sum of the distributions for each bin is identical to the input
distribution, the input distribution is still honored.

Property Modeling Petrophysical Modeling using Secondary Data • 409

 Petrophysical Modeling using Secondary Data
How to generate Crossplot bins?

Display two properties; use Upscaled cells

Create a Raw crossplot

Create crossplot bins

Delete single points, a box selection or a bin

Petrophysical Modeling using Secondary Data

Bivariate Distribution - Example of Permeability using Porosity classes

8 classes

3 classes

Modeled Permeability Modeled Permeability vs. Porosity xplot

410 • Petrophysical Modeling using Secondary Data Property Modeling

The input data have been split into three and eight different classes in
the above example. The overall correlation (after modeling) will be
much better when 8 classes are used compared with using only 3
classes.

Petrophysical Modeling Using Secondary Data –

Exercises
Examples of using secondary data will typically be using seismic
attributes (for example acoustic impedance), when modeling porosity,
or conditioning to an already modeled porosity property when modeling
permeability.
For your convenience, an extra project called Property Modeling
2010_M16.pet has been created with all necessary data in case you
have not completed any of the previous modules in the book. If you are
satisfied with your project, continue to work on it.
Porosity modeling using a Seismic Attribute as secondary
variable
In the following exercise, the seismic Acoustic Impedance (AI) attribute
is used as the secondary variable for Porosity modeling without
conditioning to a Facies model.
These variogram values used are based on the variogram analysis from
the exercise in Module 14 - Variogram analysis using correlated
attributes. The orientation, major range, and minor range values were
derived from the Acoustic Impedance data, while the nugget and
vertical range were derived from the vertical variogram analysis using
the upscaled porosity data.
Exercise Steps
1. An upscaled log named ‘PHI_Seismic(AI) (U)’ already exists
in the Exercise Model>Properties folder (it is a copy of the
‘PHI (U)’).
2. Open the Petrophysical modeling process, select ‘PHI_
Seismic(AI) (U)’ property and Sequential Gaussian
simulation as the method to use.
3. Select Zone A and define the Variogram parameters as
shown in the next figure (the parameters are from the previous
Variogram modeling).
Property Modeling Petrophysical Modeling using Secondary Data • 411
 4. Under the Co-kriging tab, select AI as Secondary variable>
Property, Collocated co-kriging as method and click the
Estimate button to calculate the correlation Coefficient.

5. Copy and Paste the settings for Zone B. In the Co-kriging

tab, re-estimate the correlation Coefficient by clicking the
Estimate button.
6. Click OK to run the model. Visually compare the ‘AI’ and
‘PHI_Seismic(AI)’ properties on a different 3D windows and
Tile them vertically to see the correlation.
7. Optionally, you can change the modeling Method for
412 • Petrophysical Modeling using Secondary Data Property Modeling
 Gaussian random function simulation for both zones.
8. Click Apply to run the model. Compare the ‘AI’ and ‘PHI_
Seismic(AI)’ properties and see the correlation.
9. In the Co-kriging tab, use the correlation Coefficient slider
bar to see the effect of changing the correlation on the result.

Permeability modeling
Typically, Permeability is closely related to Porosity. This dependency
can be included in the permeability model in different ways:
• A trend or Permeability transform that describes the correlation
between Porosity and Permeability can be imported or
generated.
• Porosity can be used as secondary input in Collocated co-
simulation.
• Porosity can be used as secondary input in Bivariate
distribution used for back-transformation of the Gaussian
Simulation methods result.
• A combination of methods 1 and 3 or 2 and 3.

Property Modeling Petrophysical Modeling using Secondary Data • 413

 Permeability modeling conditioned to a Facies model and
using Collocated co-kriging with Porosity
The Porosity model will be used as a Secondary variable to keep a
consistent correlation between Porosity and Permeability. Collocated
co-kriging will be used to honor the correlation between the two
properties.
An analysis of the correlation between Permeability and Porosity is
needed to decide whether the Porosity model can be used as Secondary
input for either Collocated co-simulation or Bivariate distribution. In a
previous exercise, the correlation between Porosity and Permeability
was analyzed. In this exercise, the correlation between the two
properties in the Petrophysical modeling process will be estimated
automatically. Both the porosity and the permeability properties will
also be conditioned to a Facies model.
Exercise Steps
1. Copy the existing property for ‘Perm’ and rename it to
‘Perm_Facies’.
2. Open the Petrophysical modeling process dialog, select the
existing property for ‘Perm_Facies’, select Zone A then the
Facies model ‘Facies_Adaptive_Model’. Select
Sequential Gaussian simulation as Method.
3. Input variograms for shale, channel, and levee facies
according to the table below:

Facies Type Major Minor Vertical Azimuth

Shale Exponential 500 500 12 0
Channel Exponential 1500 600 7 134 (-46)
Levee Spherical 1000 500 6 134 (-46)
Lobe Spherical 1000 500 6 134 (-46)

4. For each facies go to the Co-kriging tab, select ‘PHI_SGS(1)’

as the Secondary variable, select the Collocated co-
kriging option and Estimate the correlation Coefficient.

414 • Petrophysical Modeling using Secondary Data Property Modeling

 5. Copy and Paste the settings from Zone A to Zone B. Make
sure to use the same variograms for shale, channel, and levee
facies in Zone B, but remember to update the correlation
Coefficients for each facies.
6. Input the Variogram for the lobe facies as given in the table
for Zone B.
7. Click OK to run the model.
8. Check the results:
a. Visually compare the input facies model (Facies_
Adaptive_Model), the Porosity model (PHI_SGS (1)) and
the resulting ‘Perm_Facies’ property.
b. Also, compare the Histogram between the Upscaled logs
and the 3D property for each facies. Set the Properties>
Filter and look at the Histogram tab under the Settings
of ‘Perm_Facies’ property.
c. Check the correlation between ‘PHI_SGS (1)’ and
‘Perm_Facies’ with both the upscaled logs and 3D
property using the Function window.

Property Modeling Petrophysical Modeling using Secondary Data • 415

416 • Petrophysical Modeling using Secondary Data Property Modeling
Permeability modeling using Collocated co-kriging with
Porosity conditioned to a Seismic Attribute
The following exercise will use Porosity as the Secondary variable in
the Co-kriging tab when modeling permeability. However, the Porosity
property has already been conditioned to a seismic attribute property.
And by using this PHI_Seismic(AI) for conditioning when modeling
Permeability, the Permeability will also be conditioned to the seismic
attribute.
Exercise Steps
1. Copy the property ‘Perm’ and name it ‘Perm_Co-kriging’
2. Open the Petrophysical modeling process panel.
3. In the ‘Existing property’ select the property ‘Perm_Co-
kriging’ and select to model Zone A.
4. Select Gaussian random function simulation as method.
Input the Variogram using the values accoring to the next
figure:

5. Under the Co-kriging tab use the settings as shown in the

next figure. Remember to calculate the correlation Coefficient
by using the Estimate button.

Property Modeling Petrophysical Modeling using Secondary Data • 417

 6. Copy and Paste the Settings from Zone A to Zone B.
7. Re-estimate the correlation Coefficient by clicking the
Estimate button for Zone B.
8. Click Apply to run the model.
9. Check the result:
a. Visually compare the calculated ‘Perm_Co-kriging’ with
the secondary input’ Phi_Seismic(AI)’.
b. Also, compare the Histogram between the upscaled logs
and the 3D property for each facies by using the
Properties> Filter tab.
c. In a Function window check the correlation between
‘Phi_Seismic(AI)’ and ‘Perm_Co-kriging’.
Hint: If the data for the 3D model is too much to display, go to
viewport settings and remove randomly using max 1000
In the Function
points.
window the number of
points to be displayed can be
defined by using the Show

viewport settings
icon from the Function bar. In
the Settings tab>Reduce
points define the Maximun
number of points and the
Reduction type: Random,
with outliers or every n’th
point.
418 • Petrophysical Modeling using Secondary Data Property Modeling
 d. In the Petrophysical modeling>Co-kriging tab, move
the slider bar to change the correlation Coefficient and
visualize the effect on the model in the different windows.

Permeability modeling using the Porosity trend

The purpose of this exercise is to model Permeability using the different
transformations (Porosity trend) and Variogram modeling previously
performed in the Data analysis process for the Permeability property
(Perm).
The Porosity property used in Data analysis process (PHI_Data_
analysis) is a Porosity model conditioned to a Facies model.
Exercise Steps
1. Copy the ‘Perm’ property and rename it to ‘Perm_trend’.
2. Open the Petrophysical modeling process. Select the
Perm_trend property and click the Show data analysis
dialog button to see the Transformations and
Variogram modeling previously performed in the Data
analysis process. Check that they have already been done.
a. Output truncation set to truncate the permeability data
on back-transformation based on a cutoff of min 0 and
max 1300 mD.
b. Use Logarithmic transformation as the permeability data
are highly skewed.
Property Modeling Petrophysical Modeling using Secondary Data • 419
 c. The 3D trend is defined and scaled with a polynomial
function to fit the overall data and disregarding the low
perm data. Note the asymptotic endpoints to avoid the
modeled data to “take off” on extreme values.
d. The Normal Score transformation must be set to make
sure the data have a mean of 0 and [Link] of 1.

3. Click OK to close the Data analysis dialog.

4. Select either Sequential Gaussian simulation or Gaussian
random function simulation as the Method to use.
5. Click the buttons Use variogram made in data analysis
and Use the transformation made in data analysis
.

420 • Petrophysical Modeling using Secondary Data Property Modeling

 6. Click OK to run the model. Display the model and visualize the
result.

Bivariate distribution - Crossplot bins

The purpose of this exercise is to define the bivariate distribution to be
used for Permeability modeling. The bivariate distribution is, in this
case, defined by the crossplot of the permeability and the porosity
property (previously modeled using Acoustic Impedance as secondary
data).
Exercise Steps
1. Use the Function window of the previous exercise (PHI_
Seismic(AI) vs. Perm_Co-kriging). Use log transformation
for the y-axis (Permeability).
2. Click the Show upscaled well logs button to use the
upscaled log data.
3. Click the Create raw cross plot button. Enter the
Crossplot name in the dialog as ‘Crossplot PHI-Perm’. The
crossplot will be stored in the Input pane. Click the Create
cross plot bins button. Choose 3 bins (No. intervals)
and click the Create intervals button.

Property Modeling Petrophysical Modeling using Secondary Data • 421

 Once edited, the line
becomes red. Make sure to
4. Using the Select and edit Line button, you can modify
deactivate the line by the bin size interactively using the left mouse button.
clicking the top part of one of
the bins (click once to 5. The Select and edit / add points [E] button allows you
activate the whole bin, click to select individual points of the crossplot (they get highlighted
once more to deactivate it).
in red) and delete them. In this way, you can select and delete
outliers. The deleted points are not used for the CDF.
By pressing the left
mouse button and making a
selection box, you can select
several points to delete.

422 • Petrophysical Modeling using Secondary Data Property Modeling

Permeability modeling using bivariate distribution
The Bivariate distribution that was set up in the Crossplot Porosity-
Permeability in the previous exercise will be used for calculating the
porosity dependent cumulative distribution functions that are used for
the back-transformation of the Permeability simulation result.
Exercise Steps
1. Make a Copy of the ‘Perm’ property and rename it to ‘Perm_
Bivariate_3bin’.
2. Open the Petrophysical modeling process panel. Select the
property ‘Perm_bivariate_3bin’.
3. Select Zone A and Sequential Gaussian simulation as the
Method to use.
4. Go to the Distribution tab. Select Bivariate as Distribution
method. Define the settings according to the next figure.
Select the ‘From crossplot’ check box. The crossplot
‘Crossplot PHI-Perm’ is stored in the Input pane. Insert it by
using the blue arrow.

Property Modeling Petrophysical Modeling using Secondary Data • 423

 5. Go to the Variogram tab and enter the model parameters
according to the next figure:

6. Use the Copy-Paste buttons to have the same settings for

Zone B, but remember to update the Output data range in
the Distribution tab by clicking the Estimate button. Click OK
to run the model.
424 • Petrophysical Modeling using Secondary Data Property Modeling
 7. Check the result using Show property button in a
Function window with the initial crossplot.
Hint: If there are too many modeled points (red), you can
reduce the points from the Viewport settings> Settings
tab. In this case, since we are interested in the overall
distribution within each bin, in the Reduction type select the
Retain outlier option.

What can you say about the correlation of the modeled Permeability
and the Porosity? How can you improve the correlation?

Property Modeling Petrophysical Modeling using Secondary Data • 425

 8. Repeat the exercise for creating a raw crossplot. Create a
crossplot using 8 bins (No. intervals).
9. Make a copy of the property ‘Perm’ and rename it ‘Perm_
bivariate_8bin’.
10. Model the permeability using the updated crossplot with 8
bins.
11. Make a crossplot of the modeled permeability with the
porosity. This will create a better correlation between the two
properties as shown in the next figure.

426 • Petrophysical Modeling using Secondary Data Property Modeling

Summary
In this module, you have used secondary data for different approaches
to the Petrophysical modeling process. You were introduced to the
locally varying mean (LVM) algorithm. It was demonstrated that
Collocated co-kriging is helpful for correlated properties where one
property is sensitive against the input parameters and can be stabilized
via the correlation coefficient to an already existing property. It was
also shown that trends can be used to guide the population and that
Bivariate distributions help to define the target distribution function in a
more detailed way.

Property Modeling Petrophysical Modeling using Secondary Data • 427

Module 17 - Volume
Calculation and Uncertainty
Analysis

In the previous modules various logs were prepared for upscaling and
several property models were generated. These included SW, NG and
Porosity which are all essential as input to a static volume calculation.
The use of these modeled properties to perform volume calculation will
be covered in this module. Also, how to use the volume calculation as a
base case to make a simple uncertainty analysis will be shown in this
module.

Prerequisites
To successfully complete this module, the user must have knowledge of
the following:
• Familiarity with Petrophysical Fundamentals
• Familiarity with Reservoir Modeling
• Basic knowledge of Volumetrics & Uncertainty
• Petrel Introduction

Learning Objectives
The purpose of this topic is to give the participant an overview how to
use the property model to perform a volume calculation and uncertainty
analysis. At the completion of this module, you will be able to:
• Run a Volume calculation case using the modeled properties
• Use the Uncertainty and optimization process to make a simple
uncertainty analysis based on the volumetrics case

Property Modeling Volume Calculation & Uncertainty Analysis • 429

 Lesson

Volume Calculation & Uncertainty Analysis

Why create a reservoir model?

Maximize the information usage to

optimise production
Reservoir properties are critical factors
affecting production
Calculate hydrocarbon in place
Assist in the field development
Assess uncertainty

We are making big decisions based on limited data!

Volume Calculation & Uncertainty Analysis

General workflow

Facies modeling Petrophysical modeling

Make contacts Volume calculation Uncertainty and optimization

430 • Volume Calculation & Uncertainty Analysis Property Modeling

 Volume Calculation & Uncertainty Analysis
Volume calculation process

In the Volume calculation for our static model we can incorporate existing property
models to calculate Bulk volume above existing fluid contacts, and consequently
Net volume (using Ntg), Pore volume (using Porosity) and STOIIP (using SW).

Create a new Volumetric case and

select the 3D grid to use

Define the hydrocarbon intervals to use

with the Fluid Contacts created earlier

Volume Calculation
In the early stages, a Static volume calculation is applicable for getting
a reasonable overview of the oil and/or gas in place volumes (STOIIP
and GIIP). It does not replace the simulated volumes used in Dynamic
models but is suitable as a first pass. We need the structural 3D grid
as input, in addition to the main fluid contacts. As each cell is
triangulated, it is possible to cut the cell exactly, and hence use exact
contacts.
General Properties – Are properties such as NtG, Porosity,
Recoverable oil and Recoverable gas. They must exist under the
Properties folder as 3D properties. If they do not exist, then constant
values must be given.
Oil Zone Properties – Properties specific for the oil zone, for example
Formation Volume Factor of Oil (Bo), and water saturation (Sw) or oil
saturation (So) in the oil zone. If there is free gas within the oil zone,
you can enter the Sg, Bg and Rv for the free gas.
Gas Zone Properties – Properties specific for the gas zone, i.e.
Formation Volume Factor of Gas (Bg), and water saturation (Sw) or gas
saturation (Sg) in the gas zone. If there are free oil in the gas zone, you
can enter the So, Bo and Rs for the free oil.
Property Modeling Volume Calculation & Uncertainty Analysis • 431
 Volume calculation workflow:
1. Open the Utilities>Volume calculation process. Select
Create new, define the name for the new case.
2. Select the 3D Grid.
3. Define the hydrocarbon intervals to use with the Fluid
Contacts created earlier in the Corner point
gridding>Make contacts process.
4. Define Oil and Gas properties. Use either an already made
property from the 3D grid or select the constant property.
5. Under the General Properties tab, activate N/G and Porosity
properties or use constants.
6. In the Output tab, select which properties and volume height
maps to generate if needed.
7. When the Volume calculation is Run, the properties are
stored in the Properties folder of the selected grid. The
volumetrics case is stored in the Cases pane and its result in
the Results pane.

Creating a Report can be done in 3 ways:

1. Make a report while running the Volume calculation process
(Make Report button).
2. Leave it for later, run first, then right-click on the volumetrics
case run in the Cases pane and select Make volumetrics
report.
3. Double-click on Volumetrics folder in Results pane, go to
Reports tab.

432 • Volume Calculation & Uncertainty Analysis Property Modeling

 Volume Calculation & Uncertainty Analysis
Uncertainty and optimization – Process input and setup

Once a Volume calculation is run, a case is created and can be used as Base case. Specify
whether to run Uncertainty (vary all paramters) or Sensitivity (vary one parameter, holding
others constant). Define the process variables (here: $Seed)

Create a new Uncertainty run

Drop in Base case

(from Volumetrics)

Define variables for involved

parameters
Here: Poro, NtG and SW

Why Uncertainty?
In any model,. there are a tremendous amount of uncertain parameters.
It is important to establish key uncertainty parameters before launching
into a full Monte Carlo type analysis.
Some explanations of terminology:
Scenario - Represents an hypothesis involving high level deterministic
changes like a new depth converted horizon to define the top reservoir.
Realizations - Involves a variation of parameters within a hypothesis
and could be fluid contact variations, or stochastic variations of
petrophysical models (like just changing the seed number).
If there are multiple uncertain parameters in a combined run, normally a
lot of cases (realizations) should be run to cover all possible outputs
related to input distribution. To some degree, this is avoided by using
Latin Hypercube smart sampling in Petrel (default), but still multiple
runs should be performed. If only a single parameter is varying, one can
argue that just a min, max and most likely case needs to be run.
Property Modeling Volume Calculation & Uncertainty Analysis • 433
 Uncertainty - Investigates the combined uncertainties for all uncertain
parameters. In other words, assuming that all the key uncertainty
parameters are independent of each other (for example, an optimistic
value may be picked for parameter 1 while a pessimistic value may be
picked for parameter 2).
Sensitivity - A sensitivity run is useful for investigating the relative
influence of each of the uncertain parameters. The options in Petrel are:
• Sensitivity by process (involving all uncertain parameters
only in one Petrel process)
• Sensitivity by variable (concerning only one uncertain
parameter).

Uncertainty principle in Petrel

All workflows, except for the one with variables, are executed once
only. The workflow with variables is repeated in loop as many times as
the number of runs.
When running the workflow with variables, all processes and
operations defined in this workflow are executed sequentially, and then
the loop iterates again on them all.
• Establish the Base case scenario.
• Run multiple realizations (loops)
• Analyze the effect of the current variable on Volumetrics/
Simulation

Uncertainty Workflow:
1. Create a new Uncertainty and optimization run
2. Insert a Base case (Volume calculation case) from the Cases
pane.
3. Define the variables in the Variables tab, as well as in the
different processes.
4. In the Uncertainty tab, define the Number of samples
(workflow simulations).
5. Test and Run the uncertainty analysis.
The Base case is a reference case. It is used as a reference,
representing a model where all of the key uncertainty parameters can
be varied, one by one or all at the same time, resulting in a final volume
distribution. Based on the volume distribution, you may be able to pick
434 • Volume Calculation & Uncertainty Analysis Property Modeling
a P10, P50 and P90 case. Therefore, P50 will usually differ from the
Base case. This Base case model will be modified when running the
uncertainty workflows, but will be reproduced with the Base case
parameters after all sensitivities have been run.
The uncertain parameters to investigate are specified as variables in
the Variables tab (for example $SEED) and how these variables should
vary, for example their uncertainty range. You can use functions from
the drop-down list to specify how a variable should vary.
SEED variable: The constant integer value that is used as start of the
random number sequence for stochastic processes.

Volume Calculation & Uncertainty Analysis

Uncertainty – Output in Cases and Results
Once the Uncertainty is done, Petrel creates a case for each run and restores Base case. All
the results from Volumetrics (including STOIIP) are stored in Results tab.

Base case from Volumetrics

10 runs

Uncertainty outputs
All workflows, except for the one with variables, are executed once
only. The workflow with variables is repeated in a loop as many times
as the number of runs.
When running the workflow with variables, all processes and
operations defined in this workflow are executed sequentially, and then
the loop iterates again on them all.
Property Modeling Volume Calculation & Uncertainty Analysis • 435
 The schematic workflow is meant to illustrate the following: Define
a loop and define the variable(s) to be used for each loop, build the
model (and do the necessary updates to it, for example change the seed
number and compute volumes), start a new loop, etc.
After each run, the Base case must be restored again to give the same
setup as before the uncertainty was performed.
The Base case scenario is the complete Petrel model together with the
volumetrics and/or simulation performed on it.
Pre workflows – Are workflows not related directly to the uncertainty
loop; like new organizing folders etc.
Post workflows – Workflows during or after an uncertainty loop (but not
to be looped through); like calculating temporary properties, mean
values, exporting data etc.

The Cases pane>Cases work as a management system for the output:

• Used for organizing the (potentially large amount of) cases
• Can be used for Flow Simulation and Volume Calculation in
addition to uncertainty on their results (need one start/base
case)
• Possible to use a case collection to combine cases into one
folder
Petrel will run the specified amount of runs and display each case in
the Cases pane. All parametric results are stored in the Results pane.

436 • Volume Calculation & Uncertainty Analysis Property Modeling

 Volume Calculation & Uncertainty Analysis
Uncertainty – Display of P10, P50 & P90
Select cases from Cases tab and parameters of interest (here STOIIP) from Results tab in a
Histogram window. Petrel displays P10, P50, P90. To show the P50 case and related Seed
number, use the Variable Spreadsheet, sort on STOIIP and use %Percent ranks.

P10 P50 P90

STOIIP

P50 case $Seed

Uncertainty results and display

Petrel will run the specified amount of runs and display each case in the
Cases pane. The parameters set up in the Volume calculation
process are stored in the Results pane.
Display in Histogram
The Histogram window allows you to view a selection of cases and
their corresponding volumetric results. It will also compute the P10, P50
and P90 based on the distribution. However, you cannot find the exact
case number representing your P10, P50 or P90 case.
The histogram can display, for example the STOIIP (from the Results
pane) with all cases (from the Cases pane). Then, you can split the
histogram based on Zones or Segments using a right-click option.
Variable spreadsheet
All variables used in the Uncertainty run are stored in a Variable
spreadsheet. This can be launched from a case or the Uncertainty
case folder as a right-click option.

Property Modeling Volume Calculation & Uncertainty Analysis • 437

 To view the variation in STOIIP for each case, select it from the Show
volumetric button drop-down menu. Once shown in the spreadsheet,
highlight it and click the %Percent ranks button. Petrel will now sort
the cases according to rank and the closest case to P50 is highlighted in
The P50 value shown green. The corresponding seed number and case can now easily be
in the histogram does not found. By right-clicking on the corresponding case in the Cases pane, a
necessarily match the P50
case from the Variables
Create workflow and Rerun workflow can be performed.
spreadsheet.

Extra: Workflow editor and Uncertainty Analysis course

(2 days)

• Workflow editor interface

• Processes and Operations
• Pre-defined workflows
• Instantly update your model with
new input data
• Uncertainty and optimization
• Integrated structural uncertainty

438 • Volume Calculation & Uncertainty Analysis Property Modeling

Volume Calculation & Uncertainty Analysis -
Exercises
Volumes are calculated in the Volume calculation process. A pre-
defined Contact set in the Make contacts process is used as input to
the Volume calculation process.
You will specify the input to be used and the type of output to generate
(3D properties, reports, maps etc.). If you want to see the effect of
different property models, then you should define several Volumetric
runs, each using different inputs; such as several realizations of
properties or contacts. A report defined by you will be created after the
process has been run, listing all of the volumes per zone, per segment
and/or per facies.
Volume calculations can be performed using several hypotheses (Runs)
in one operation. Running a volume calculation will create a case,
stored in the Cases pane. The Results pane controls outputs and
allows you to create filters for separating out data and to show the
results of the run in a plot window.
This exercise covers how to run a volume calculation using the
petrophysical property models we have made previously (SW, NtG and
Porosity) and the use of the volume calculation case to perform a simple
uncertainty analysis.
For your convenience, an extra project called Property Modeling
2010_M17.pet has been created with all necessary data in case you
have not completed any of the previous modules in the book. If you are
satisfied with your project, continue to work on it.
Volume calculation using the property model
Exercise Steps
1. Open the Processes pane>Utilities>Volume calculation
process. Select Create new case and enter a name (for
example, ‘Case 1’). Use the Exercise Model as Grid.
2. In the Properties>Fluid zones tab, select only the Oil
Hydrocarbon interval.
3. Insert the Oil water contact by selecting it from Exercise
Model > Fluid contacts > Contact set > Oil water contact
and clicking the blue button in the Contacts tab (see figure
below).
Property Modeling Volume Calculation & Uncertainty Analysis • 439
 4. In the Properties>General tab, specify the N/G (NtG_cont_
synt(1)(U)) and Porosity (PHI_SGS(1)(U)) properties from your
model. Note that all of the properties that you select should be
populated in the 3D grid previously!

5. In the Properties>Oil tab, specify the Sw property (SW_

Archie), set Bo to 1.19 and leave all the other constants as
default.

440 • Volume Calculation & Uncertainty Analysis Property Modeling

 6. Optionally, in the Settings>Output tab, you can specify which
properties and volume maps you would like to generate, as
well as how you want to set up the Report format.
a. Select Make spreadsheet report.
b. Select only HCPV oil and STOIIP as the properties to
make.
c. Click the Report settings button, go to the Report
tab>Cases tab, select only Bulk, Net and Pore Volume
in addition to HCPV Oil and STOIIP. In the Format tab
select only Zone and OK.
7. Leave all the other settings as default, click the Apply and
Run buttons to perform the volume calculation.
8. Check the volume report in the Output sheet. The new volume
calculation case is stored in the Cases pane.
You can run several cases using the different property models (NtG,
Porosity and Sw) or typing a constant value for the properties. By
inspecting the results, you can understand how the various changes
impact the volumetric result.

Uncertainty analysis using the volume calculation case

For this simple case of uncertainty, we will just vary the SEED number
for the NtG, Porosity and Sw models in the Petrophysical modeling
process and define the number of realizations as, for example, 10. We
want to have a look at how much the STOIIP varies if the properties
vary, or simply compare different case outputs. The Uncertainty process
will generate 10 volume calculation cases with 10 different realizations
of NtG, Porosity and Sw properties.
Exercise Steps
1. Open the Processes pane>Utilities>Uncertainty and
optimization process. Select Create new case and enter a
name (for example, use the default name ‘Uncertainty and
optimization 1’).
2. In the Base case tab, insert the case by selecting the volume
calculation case from the Cases pane by clicking the blue
arrow button. In the process dialog, all processes involved in
the 3D grid (depth) building and volume calculation case will
appear.

Property Modeling Volume Calculation & Uncertainty Analysis • 441

 3. In the Uncertainty and optimization process > Base case
tab, double-click on the Petrophysical modeling processes
for the NtG, Porosity and Sw properties to define a global
SEED variable as $SEED. Click OK to close the processes.

4. After the $SEED variable is defined, the $SEED variable will

appear in the workflow window along with the process where
it has been declared, as shown in the figure below.

442 • Volume Calculation & Uncertainty Analysis Property Modeling

 5. In the Variables tab, enter a value for the Seed variable:

6. In the Uncertainty tab, define the No. of samples as 10, click

the Apply, Test and Run buttons.

7. Go to the Cases pane, a new folder is generated called

‘Uncertainty and optimization 1’. In this folder, there are 10
volume calculation cases from the uncertainty analysis.
8. Open a new Histogram window and select the 10 cases to
compare.
9. Go to the Result pane and select the STOIIP property.

Property Modeling Volume Calculation & Uncertainty Analysis • 443

 10. In the Histogram window>function bar, click the Show cdf

curve button to show the cumulative distribution curve

in the histogram. In the Histogram window you can visualize
the P10, P50 and P90 cases.
11. Go to the Cases pane>Uncertainty and optimization 1
folder > right-click and select the option Show variables
spreadsheet. In this new window, you can see the variable
values and results for each case. Click the Show volumetrics
button and select STOIIP; it will be added to the table.

12. Click the STOIIP column in the spreadsheet to highlight it

(gray) then click the %Percent ranks button to rank the cases.

The case closest to P50 is highlighted in green. The smallest

value does not correspond to 50% because the computation
assumes a distribution, not just a discrete set.

444 • Volume Calculation & Uncertainty Analysis Property Modeling

Use of realizations
Petrophysical modeling realizations
Earlier we looked at generating realizations directly from the
Petrophysical modeling process. This will create multiple, equally
probable outputs (with unique seed numbers) which are stored in the
Properties folder (suffixed with realization numbers). However, these
are not necessarily used further but can be suitable as a visual check
for areas of high/low uncertainty. The arithmetic mean can give you
some idea of the trend/highest accumulation in the data, but the mean
has no seed number to be re-generated, so it cannot be used as a Base
case in an Uncertainty analysis.
Uncertainty realizations:
In the Uncertainty process, the seed number can be varied and we get
a direct measure of the impact of the seed on volume results (other
parameters can of course also be varied). If the uncertainty is low and
we have good well coverage, the variation in volume will not be too
high. If only a few wells are available, the impact of varying the seed
will have a bigger impact on the volumes. The P50 case can easily be
found and also the P10 and P90 cases can be found and re-generated.
The physical property can be generated and visualized through the
automatically set up workflow.
In the end, the Reservoir Engineers may want a P50 case to work with
in Prediction or History matching, and they can in turn create their own
realizations in for example the ECLIPSE simulator (also integrated in
Petrel).

Property Modeling Volume Calculation & Uncertainty Analysis • 445

 Summary
In this module, the Property models generated throughout the course
were used to give a numerical input for a simple static volume
calculation, and to assess uncertainty. Also, a short introduction to the
Volume calculation and Uncertainty and optimization processes
were given.

446 • Volume Calculation & Uncertainty Analysis Property Modeling

Module 18 - Case Study

Porosity Modeling using an Acoustic Impedance

cube as secondary information
This module covers the workflow of how to use an Acoustic Impedance
cube as secondary data for Property modeling. We will use a different
data set, including seismic data, with an Acoustic Impedance cube. The
exercises will also cover a testing phase using various 3D grids with
different resolutions to test the relationship between Impedance log
(high frequency) and resampled seismic (low frequency).

Prerequisites
To successfully complete this module, the user must have knowledge of
the following:
• Familiarity with Geostatistics Fundamentals
• Familiarity with Geophysics Fundamentals
• Familiarity with Reservoir Modeling
• Petrel Introduction

Learning Objectives
The purpose of this topic is to give the participant a general
understanding of how to use an Acoustic Impedance Cube as a
secondary data for porosity modeling. At the completion of this training,
you will be able to:
• Generate new 3D grids by copying over relevant zones, re-layer
and re-upscale data.
• Use Acoustic Impedance to correlate for possible Co-kriging/
Simulation and quality check
• Make Porosity models based on porosity logs and seismic
acoustic impedance information

Property Modeling Case Study • 447

 Lesson

Case Study
Initial Analysis

Problem:
Primary attribute not sampled in the entire 3D grid.

Solution:
1. Check whether seismic acoustic impedance cube correlates with
primary attribute.
2. Extract the variogram model parameters from secondary attribute
3. Use secondary attribute in collocated co-simulation

Case Study
Check Relationship between Impedance log and the Property
Basic assumption:

Does a relationship exists between the Log (Imp) and the Seismic
property (AI)? This needs to be tested:

For each well calculate an impedance log from sonic and density logs
Make a crossplot between the impedance log and the reservoir property log

⇒ If no density log is available use the sonic log instead of the impedance log for the
crossplot
⇒ If this test fails the AI cube cannot be used as secondary input for the property
modeling

448 • Case Study Property Modeling

 Case Study
Crossplot Impedance log vs. Porosity log
A negative correlation exist from the crossplot, but seems to be weaker for low
porosity values. Impedance can be used for modeling upper zones only.

Good correlation
Note:
For low porosity values the
Bad correlation impedance is less sensitive.

Case Study
Crossplot Impedance Log vs. VCL Log
A general low correlation exist, indicating that an acoustic Impedance cube may not
be valid for Shale modeling

Bad correlation

Note:
No relationship between
impedance and amount of shale.

Property Modeling Case Study • 449

 Case Study
Limited Seismic Resolution

Resolution of seismic AI:

In the range of 15-30m.

Resolution of impedance log:

1m and less

Low impedance bands (red) are

about 1-3m thick

Case Study
Resampling of AI Cube into 3D Grid

The layer thickness of the model should be guided by the vertical

variogram range of the log data.

However the layer thickness may have an effect on the relationship

between the resampled AI and the upscaled log data (introduction of
‘noise’ with decreasing layer thickness).

This needs to be checked!

450 • Case Study Property Modeling

 Case Study
Influence of Layer thickness
Upscaled Imp log – Resampled AI relationship

Layer Thickness:
5m
3m
1m

Resampled Seismic AI

Smaller layer thickness reduces correlation

Case Study
Influence of Layer thickness
Upscaled PHI log – Resampled AI relationship
Layer Thickness:
5m
3m
1m

Resampled Seismic AI

Smaller layer thickness reduces correlation

Property Modeling Case Study • 451

 Case Study
Seismic AI and Porosity Relationship by Zone

Top reservoir Bottom reservoir

Bad correlation
Ok correlation

Resampled Seismic AI

Seismic AI cannot be used for porosity modeling in Bottom reservoir

Case Study
Horizontal Variograms from Seismic AI

1. Calculate a variogram map from seismic

acoustic impedance to examine anisotropy
2. Compute sample variograms along the
major and minor direction
3. Fit the model to the experimental variogram
4. Use these variogram parameters for porosity
modeling
5. Justification:
Areal distribution of log data is usually very sparse
Dense correlated attributes can provide a better
quantitative estimate of spatial correlation

Seismic AI shows
Zonal Anisotropy

452 • Case Study Property Modeling

 Case Study
Crossplot Seismic AI vs. Modeled Porosity
Porosity model without secondary input Porosity model with seismic AI as secondary input

Red: AI - Modeled Porosity

Black: AI – Upscaled Porosity

Case Study
Porosity Modeling - Seismic AI as Secondary Input

Acoustic Impedance Porosity Porosity

Resampled Seismic No secondary Attribute Collocated Co-kriging
with AI (CC: -0.54)

Property Modeling Case Study • 453

 Porosity Modeling using an Acoustic Impedance
Cube as Secondary Information - Exercises

Become familiar with the Acoustic impedance cube

This exercise covers the necessary steps for using a seismic acoustic
impedance cube for reservoir property modeling. Scroll through the
volume using inlines, crosslines and time slices. Display relevant logs.
Exercise Steps
1. Open the project Acoustic Impedance [Link].
2. In a 3D window display a section (Inline or X-line) of the
Acoustic Impedance cube (AI (Realized)). It is stored in the
folder Seismic>Depth Seismic in the Input pane.
3. Display the wells and the Impedance log in the 3D window.
Compare them to the sections of the impedance cube that are
close to the wells. To see the logs more easily against the
Intersections, draw log as 3D pipe (log>Style tab).
4. Open the 3D grid GEO1 Reference. Right-click on the
Intersections folder and create a General Intersection,
click the blue button in the lower left part of the Petrel
interface and select the Horizons and Faults to be displayed.
Scroll through the model.
5. The reservoir is given by the zones ‘lem’, ‘aeo’ and ‘wat’.
6. The Acoustic Impedance cube is already sampled into this
model (the property Al (realized)). Also, display this on the
General Intersection plane and scroll through the model. Click
the button Align plane horizontally at the bottom of
the Petrel interface and scroll through the model. You will
quickly spot the sandy (red) parts of the reservoir (lem-aeo-
wat).

454 • Case Study Property Modeling

Calculate an acoustic impedance log for each well
An acoustic impedance log (imp) to be used in the following exercise
will be calculated using the well log calculator. The calculation is
performed as a global log for all three wells, using 1/Sonic (as velocity),
Density and multiplicators for converting microseconds to seconds and
feet to meters.
Exercise Steps
1. In the Input pane, open the Wells folder and right-click
Global well logs to open the Calculator. Always check the
2. Define the settings according to the figure: project units against the
various log units (templates)
when doing these kind of
calculations. In this project,
the project settings are
Metric (meters). The Sonic
log is in microseconds/feet
and density is in g/cc. Since
the project is metric, Petrel
will choose a metric standard
for the Acoustic Impedance
template (kPa.s/m).

Property Modeling Case Study • 455

 Cross plot impedance log against porosity log
The following step is necessary to determine whether there is a
relationship between the acoustic impedance log and the porosity log.
Exercise Steps
1. Open a Function window.
2. In the Wells folder, select the ‘phi_edt’ log (x-axis) and the
new ‘Imp’ log (y-axis) from the Global well logs folder.
Select all of the three wells to compare the difference.
3. Use the Log transform button for the y-axis (in the
function bar).
4. Check whether there is a (linear) relationship between these
two logs. Compare the relationship for high and low impedance
values.

5. Make a crossplot between the ‘VCL1’ (shale) log and the ‘Imp’
log and check for a correlation using the Make Linear
Function from Crossplot button . This can only be done
for one well at a time unless you make a crossplot using the
Create Raw Crossplot button .

456 • Case Study Property Modeling

Check the model resolution
In the following exercise, we will assume that we have just finished the
structural model and have begun working on the property modeling. To
set up this situation we copy the structural part of the finalized 3D Grid
“GEO1 Reference” into a new 3D Grid and give it the name
‘Training’.
Exercise Steps
1. In the 3D grid GEO1 Reference go to Zones filter and
deselect the zone ‘rot’.
2. Open the Settings of GEO1 Reference. Select the Output tab
and set the parameters accoring to the figure below. Click the
Copy global grid button. A new 3D grid is made called ‘Part
of GEO 1 Reference’.

Property Modeling Case Study • 457

 3. Rename the copied 3D grid to ‘Training’ and make sure it is
active.
4. In the Layering process (Corner point gridding process
folder), calculate layers of 1 meter cell thickness.

5. Delete all properties in the Property folder of the 3D grid

‘Training’. They were corrupted during the process of layer
calculation (since the layers were updated then the properties
must be updated as well).
6. In the Scale up well logs process, upscale the logs ‘PHIE’,
‘PHIT’ and ‘Imp’.
7. Activate the Data analysis process and select the upscaled
property ‘PHIE’ and deselect the Zones. The only
transformation that is needed is a Normal Score
transformation.
458 • Case Study Property Modeling
 8. Go to the Variograms tab and select the Vertical direction
tab. The vertical analysis should first be done on raw log data (
button) and the Simbox button should be turned off.
Estimate the vertical range. Note that Simbox mode is off, but
it should be turned on once the variogram has been
established.
9. Make sure the Lag distance of the Search cone is set
appropriate to sampling of logs.

Property Modeling Case Study • 459

 Note: In this project, the wells have some deviation. This will affect the
vertical variogram calculation; that is, the Band width etc. now have
an impact compared to a vertical wells where one only need to adjust
Search radius and Number of lags.

Create layers with thickness derived from data analysis

Initially, we had a grid with 1m layering but performed a variogram test
and will now optimize the grid using 3m layering.
Exercise Steps
1. According to the vertical range from the Data analysis
A layer thickness of process, the layers thickness should be approximately 3 m
3 m is chosen here due to
Vertical variogram analysis (6/2=3).
and in order to do a fast 2. Reopen the Layering process. Make the cell thickness 3 m for
computation when each zone.
resampling the acoustic
impedance cube into the
model.

460 • Case Study Property Modeling

Resample the property logs and acoustic impedance cube
into the model
Since the layering was redone, the properties must also be redone to
average correctly in the new cells.
Exercise Steps
1. Make sure the 3D grid ‘Training’ is active.
2. Open the Scale up well logs process.
3. Re-upscale the logs ‘PHIE’, ‘PHIT’ and ‘Imp’.
4. Resample the acoustic impedance cube: Under the Property
modeling processes folder, select the Geometrical
modeling process using Seismic resampling. Choose the
parameters as shown in the next figure.

5. Display the sampled cube in the 3D Window. Use the I, J, and

K-filter for scrolling through the cube.
6. Rename the new property to ‘AI’ (as it takes a default name
from the template during resampling).

Property Modeling Case Study • 461

 Cross plot the resampled seismic cube with the
impedance upscaled log
This exercise shows that the relationship between the seismic acoustic
impedance and the resampled log impedance is influenced by the
chosen layer thickness The project contains grids with a layer resolution
of 1m and 5m in addition to the 3m resolution model that you have
created. The relationship between the seismic impedance and the log
impedance is checked from crossplots.
Exercise Steps
1. Open a Function window and crossplot the resampled
seismic impedance cube ‘AI’ against the upscaled log
impedance ‘Imp’ for the three models GEO 1m, Training and
GEO 5m. The upscaled logs can be found in the Properties
folder of the models. All crossplots should be displayed in the
same Function window.
2. For the above-mentioned 3D grids, right click on the
Properties folder for each of them and open the Settings. In
the Style tab, select a different Specified color for the
This coloring must be property. This allows you to distinguish between the data of
performed since we are the different models.
displaying multiple data from
different grids.

3. Analyze whether the layer resolution has an influence on the

relationship between the seismic ‘AI’ and log ’Imp’.
4. Switch off all crossplot data. Crossplot the seismic ‘AI’ against
the log ‘Imp’ for the Training model (with 3m layer resolution).
5. Use the Zone filter to find out whether some of the zones
show a better match than the others.
6. Open the Geometrical modeling process. Select the method
Zone index (from all zones) and click OK.

462 • Case Study Property Modeling

 7. Display the created Zones index property as ‘Z-value’ in the
‘Crossplot AI vs Imp’. It will color code the crossplot
according to the zones the data belongs to as shown in the
next figure.
8. Click on the Show/Hide auto legend button in the top
toolbar to display the zones symbol legend.
9. Decide which zones can be used for the seismic guided
porosity modeling (by ‘good enough correlation’).
Note: If you do not get the full scale view of the data, there may be
outlier values in the resampled AI property. To fix the scale, first inspect
the histogram for the AI property, then go to the settings for the
Function window and adjust the scale (here X-axis).

Property Modeling Case Study • 463

 Derive the variogram model parameters from the acoustic
impedance property
Variogram analysis is used to determine the spatial variation for a given
property. In cases where sparse or very little data is present, horizontal
variograms must frequently be implied from geological knowledge.
However, a densely sampled correlated attribute (like the AI cube) can
serve as a substitute for horizontal variogram analysis.
A recommended procedure is to compute a variogram map first. This
map can then be displayed and examined to determine if sufficient
anisotropy is present. If present, the direction can be measured directly
on the map.

464 • Case Study Property Modeling

Once the principal axis direction has been determined, a sample
variogram can be computed for the major and minor direction. The
resulting variogram values can then be entered in Petrophysical
modeling or Data analysis process dialogs.
Variogram maps can be calculated separately for individual zones. Use
the zone filter to select the zone that shall be used for the Variogram
map calculation.
Exercise Steps
1. Display the ‘AI’ property in a 3D window.
2. Go to Zone Filter in the 3D grid ‘Training’ and deselect the
Zone ‘wat’ (no correlation of Seismic acoustic impedance ‘AI’
with upscaled log impedance ‘Imp’).
3. In the property Filter, apply the Index filter as shown below.

It can be beneficial to
use a property filter to
decimate the AI property
4. After selecting the ‘AI’ property, right-click to access the during the calculation. This
Settings dialog. Select the Variogram tab and select the will help speed up the
computation without
Horizontal variogram map option according to the figure: decimating the AI data too
much. In the variogram tab,
toggle the Use property
filter. We recommend using
index filtering parameters to
limit the variogram map
calculation.

Property Modeling Case Study • 465

 5. In the XY range tab, set the search distance for the X and Y
direction to be 5000 m as shown in the figure above.
6. On the Z range tab, use None as the vertical search distance.
7. Make the variogram map by clicking the Run button.
The variogram map 8. Display the variogram map computed from the ‘AI’ property in a
object will appear under the Map window. It is stored in a new folder called Variograms
Variograms folder of the
active 3D grid.
created under the Training 3D grid. Click the View all button
to make the complete variogram map visible. Adjust the
contour interval to reveal the anisotropy.

466 • Case Study Property Modeling

 9. Using the measure distance button , place a line on the
map following the major trend axis and record the orientation /
azimuth reported in the status bar located beneath the map
window.
10. Return to the Variogram tab and select the Sample
variogram: Horizontally option.
11. On the Orientation tab, deselect the isotropic variogram
option and adjust direction to match the value recorded in the
previous step. This will allow you to compute an experimental
variogram for the major axis direction. Adjust the angular
Tolerance to approximately 30 degrees.

Property Modeling Case Study • 467

 12. On the XY range tab, set the number of lags to 15 and use
Horizontal search radius of 5000.
13. On the Z range tab, use None as the vertical search distance.
14. Run the variogram computation. When completed, a variogram
model will appear under the Variograms folder of the 3D grid.
15. Display the Sample Variogram in a Function window. Make
sure that the range can be clearly seen.
16. Change the Horizontal search radius to 10000 and the
Number of lags to 30. Select the check box Overwrite last
and click Run. The display will automatically be updated. Click
the View All button to make the complete variogram
visible.
17. Deselect the Overwrite last check box and repeat the same
procedure after adjusting the orientation to 90 degrees. This
will allow you to compute the experimental variogram for the
minor axis direction.
468 • Case Study Property Modeling
 18. Display the experimental variograms for the major and minor
directions. Decide on the type of anisotropy.
19. Insert a new variogram object (Right-click>Insert
variogram under the Variograms folder located on the Zonal anisotropy can
be modeled like an isotropic
Models pane or use the Insert variogram button ). Variogram. Use one of the
Rename this new variogram object to “Var. from AI”. sample variograms for
defining the range and the
nugget. Remember that the
sill has no influence on the
simulation/kriging result.

20. In the Style tab of ‘Var. from AI’, deselect Show in min
direction of the variogram curve. The data set shows zonal
anisotropy, therefore, you need only one variogram model for
both horizontal directions (isotropy).
21. Display the Variogram model in the Function window along
with the Sample variograms for the major and minor axes.
22. At this point, adjust the point representing the (range, sill and
nugget) value for the variogram model to better fit the
experimental variogram data. This can be done graphically by
first selecting the Select and edit line button and then
editing the point in the Function window (You can also use the
‘Min’ and ‘Max’ range text areas located on the Settings tab
of the Variogram Settings dialog).
23. The Variogram model parameters vertical range and nugget
are calculated form the porosity log using Data analysis. You
did it in the exercise “Check the model resolution”.
Property Modeling Case Study • 469
 Porosity modeling based on seismic acoustic impedance
Calculate different porosity models with the seismic acoustic
impedance as secondary input in Sequential Gaussian Simulation.
Exercise Steps
1. Copy-Paste the property ‘PHIE’ for the Training grid, and
rename it to ‘PHIE Reference’. Display it in a 3D window
and makes sure the Index filter is turned off.
2. Open Petrophysical modeling process and select ‘PHIE
Reference’.
3. Run the porosity model using SGS for zones lem, aeo and
wat. Use the horizontal and vertical Variogram model
parameters extracted in the previous exercises.

Horizontal Vertical Nugget Model type

Range range
3500 5 0.1 Spherical

4. Copy-Paste ‘PHIE Reference’ and rename it to ‘PHIE

Co-kriging’. Select the zone lem. Keep the same Variogram
model parameters as for ‘PHIE Reference’. Go to the Co-
kriging tab and select the property ‘AI’ property. Under
Method section, select the check box for Collocated Co-
kriging and click the Estimate button to calculated the
correlation Coefficient.

470 • Case Study Property Modeling

 5. Click OK to run the model.
6. Compare the result with ‘PHIE Reference’ and with the
seismic acoustic impedance property ‘AI’.
7. Open a Function window and set up the crossplot of seismic
‘AI’ against ‘PHIE Reference’. Compare the crossplot with a
crossplot ‘AI’ vs. ‘PHIE Co-kriging’.

Property Modeling Case Study • 471

 Summary
In this module, a complete workflow of seismic attribute
implementation was accomplished. Seismic resolution issues were
addressed and treated in the correct way. After the seismic resampling
into the 3D grid, porosity was modeled under strong guidance of the
correlated seismic attribute using Collocated co-kriging.

472 • Case Study Property Modeling

References

GSLIB Geostatistical Software Library and User’s guide

Applied Geostatistics Series, Oxford University Press 1998
Clayton V. Deutsch and Andre G. Journel

Geostatistical Reservoir Modeling

Applied Geostatistics Series, Oxford University Press 2002
Clayton V. Deutsch 

An Introduction to Applied Geostatistics
Oxford University Press 1989
Edward H. Isaaks and R. Mohan Srivastava

Seismic Reservoir Characterization

- An Earth modeling Perpective
EAGE Publication 2007
Doyen, P.M.

Property Modeling References • 473

Index
Zonal, 68
Applied Variogram Modeling, 58
Archie’s equation
for water saturation, 145
Artificial Neural Network, 139
Automatic Legend
Symbols defined, 24

1D filters, 45 Averaging methods

for upscaling continuous logs, 172
2D filters, 45 for upscaling discrete logs, 170
3D grid
defined, 27 B
3D grids
Bandwidth
right sizing, 203
defined, 64
Base Case
A defined, 434
Acoustic Impedance Bivariate distribution
creating a variogram map from an acoustic impedance how to define and use for permeability modeling, 421
property, 373 used in back-transformation, 408
creating sample variogram from an Acoustic Impedance
property, 376 Body index, 274

Acronyms
used in this manual, 27
C
Adaptive channel modeling, 268 Calculating
body index, 274 variogram map of point data set, 73
channel connectivity control, 273
for zone A, 282 Carbonate depositional environments, 218
for zone A using a body index property, 290 CDF curve
for zone A using trends, 288 displaying and analyzing, 46
geometry, 275
layer tolerance, 272 Change of support effect, 108
relative sinuosity, 275 affine correction, 109
setup, 271 Class, 35
trends, 276
Classical Variogram, 63
Affine correction, 109
Co-kriging, 388, 403
Aggradation angle, 312
Collocated co-kriging, 404
Algorithms characteristics, 404
for Kriging, 81 permeability conditioned to a facies model and modeled using
gaussian random function simulation, 115 Collocated co-kriging with porosity, 414
sequential gaussian simulation, 102, 115 permeability modeling using collocated co-kriging with
Angle Tolerance, 64 porosity contitioned to a seismic attribute, 417

Anisotropy, 32, 62 Conceptual model, 22

defined, 24 Continuous
Geometric Anisotropy, 68 defined, 24
Property Modeling Index • 475
 Continuous well logs Deterministic Algorithms
averaging methods for upscaling, 172 defined, 24
bias to a discrete log, 173
Deterministic model, 22
upscaling, 172
Deterministic techniques
Corner point grid
assign values, 125, 129
defined, 24
closest (estimation), 129
Correlation, 32 functional (estimation), 129
defined, 24 in facies modeling, 124
degree of, 43 in petrophysical modeling, 128, 323
negative, 43 interactive, 125
positive, 43 moving average (interpolation), 129
neural net, 129
Correlation analysis, 44, 199
Discrete
Correlation coefficient, 44, 405
defined, 24
displaying a crossplot and calculate the correlation
coefficient, 49 Discrete logs
averaging methods for upscaling, 170
Correlation line, 200
bias, 173
Cox Box upscaling, 170
defined, 330 weighting, 171

Crossplot Distribution
defined, 24 creating a Normal Distribution from a Histogram, 48
Standard Deviation, 39
Crossplots variance, 39
defined, 43
displaying a crossplot and calculate the correlation Distribution shapes, 37
coefficient, 49 gamma distribution, 38
for bivariate analysis, 200 lognormal distribution, 38
normal distribution, 38
Cumulative Distribution Functions, 36, 409
Cumulative Histogram, 35 E
Cyclicity, 59
Estimation, 101

D Experimental Variogram, 57
Exponential, 57
Data analysis
defined, 24
F
Data Transformation
applying normal score transformation, 341 Facies logs
applying trend transformation, 343 calculating, 151
cox box, 330 how to setup a neural net to create a facies log, 139
distribution range, 328 interactive interpretation, 152
distribution shape, 329 preparing for interpretation, 136
for petrophysical modeling, 325
logarithmic, 329 Facies modeling
truncating distribution range, 328 3D probability, 230
adaptive channel modeling, 268
Degree of Correlation, 43 data analysis, 227
Depositional environment, 225 defined, 25, 216
deterministic algorithms, 124
facies probability analysis, 238
476 • Index Property Modeling
 facies thickness analysis, 236 Geometric Anisotropy, 68
indicator variogram, 231
Geostatistics, 30
methods, 223
object based, 222 Global fraction, 250, 304
object modeling, 267 influence on facies modeling, 254
pixel based, 222
Sequential indicator simulation, 245 GSLib
stochastic algorithms, 126 Sequential Gaussian simulation, 102
thickness histogram, 229
variogram analysis, 239 H
vertical facies proportion analysis, 235
vertical proportion, 228 Hierarchical facies modeling, 281
well correlation, 224 for zone A, 295
Fault Analysis Hierarchical modeling, 281
defined, 25
Histograms
Filters, 45 analyzing and displaying, 46
creating a normal distribution, 48
G defined, 25, 34
for continuous properties, 196
Gamma distribution, 38 for discrete properties, 195
parameters, 35
Gas Zone Properties typical uses, 35
defined, 431 visualizing, 36
Gaussian, 58 Horizon in a 3D Grid
defined, 25
Gaussian random function simulation, 354
defined, 115 Horizontal variogram maps, 63, 364
porosity modeling based on a facies model, 370
Gaussian random function simulation (stochastic), 130 I
Gaussian simulation
Icons
anisotropy influence on variogram, 111
used in this manual, 16
change of support effect, 109
compared to kriging, 101 Indicator kriging, 125
comparing result to input, 107
comparing result to kriging, 107 Indicator variogram, 231
gaussian random function simulation, 115 Input truncation, 328
influence of the variogram parameters, 110
nugget influence on variogram, 111 Interactive Petrophysics plugin, 136
principles of, 105
Intersection
unconditional simulation, 113
defined, 25
variogram model, 110
variogram nugget, 111
variogram range, 110 K
Gaussian simulation algorithms
Kriging
local varying mean, 401
algorithm, 81
General Properties as estimation technique, 86
defined, 431 calculating variance, 82
compared to gaussian simulation results, 108
Geometrical modeling defined, 25, 80, 93, 386
defined, 25 expert settings, 390

Property Modeling Index • 477

 modeling a SW property, 390 Monte Carlo Simulation, 31
modeling a SW property using kriging method and trend data,
Multi-point facies simulation, 127
396
ordinary, defined, 85
principles, 82 N
simple, defined, 84
Negative Correlation, 43
Kriging algorithms
in Petrel, 386 Net to gross
calculating a continuous log, 150
Kriging by Gslib
creating a continuous NtG log, 162
defined, 93, 386
creating a discrete NtG log, 162
Kriging by Gslib (estimation), 128 upscaling and modeling, 174
Kriging (estimation), 128 Neural networks, 125, 129, 139
classifications, 153
Kriging Interpolation how to setup a neural net to create a facies log, 139
defined, 93, 386 using nn to make well logs, 158
Kriging interpolation (estimation), 128 Nodes
defined, 25
L Normal distribution, 38
defined, 38
Lag, 53
Normal score, 331
Lag tolerance, 65
Normal score transformation, 356
Layers
defined, 41, 42
3D grid right sizing, 203
correcting right-sizing, 205 Nugget, 53
how to check the layer resolution, 206 influence on variogram model, 111
in adaptive channels modeling, 272 values for variograms, 92
using vertcal variogram to select thickness, 201
Layer tolerance, 272 O
Linear Equation System, 87 Object based
Local Varying Mean, 401 facies modeling, 222
transform using input distribution, 402 Object facies modeling
units, 402 adaptive channel modeling, 268
Logarithmic, 63 channel facies and isolated object facies modeling for zone B,
distribution shape, 329 285
geometry and distribution, 280
Logical filters, 45 hierarchical modeling, 281
Lognormal distribution, 38 overview, 267
probability trends, 280
process, 278
M
Object modeling, 126
Median, 35 overview, 267

Mode, 35 Oil Zone Properties

defined, 431
Model
defined, 25 OpenSpirit, 136
Ordinary Kriging, 85
478 • Index Property Modeling
Output truncation, 329 permeability conditioned to a Facies model and modeled using
collocated co-kriging with porosity, 414
permeability modeling using 3D trend, 407
P permeability modeling using collocated co-kriging with
porosity contitioned to a seismic attribute, 417
Pairwise Relative, 63
porosity modeling using a seismic attribute as secondary
Permeability modeling, 413 variable, 411
defining bivariate distribution, 421
Pixel based
using bivariate distribution, 423
facies modeling, 222
using the porosity trend, 419
Porosity modeling
Petrophysical logs
using a Seismic Attribute as secondary variable, 411
preparing for interpretation, 144
Positive Correlation, 43
Petrophysical modeling
applying normal score transformation, 341 Primary variables, 43
applying trend transformation, 343
correlation, 326 Probability, 32
correlation with a secondary data, 371 defined, 25
co-simulation with secondary variables, 361 of Normal Distribution, 39
creating a variogram map from an acoustic impedance Properties
property, 373 making a variance output property, 84
creating a variogram model based on the sample variogram, stationarity, 33
378
creating sample variogram from an Acoustic Impedance Property Filters
property, 376 for univariate analysis, 197
data analysis process for permeability using a porosity trend, Property modeling
349 defined, 23, 25
data transformation, 325, 340
defined, 25
deterministic techniques, 323 Q
directional variogram analysis for porosity, 347
horizontal variogram, 338 Quality check
normal score transformation for gaussian simulation, 356 correlation, 199
overview, 322 editing correlation line, 200
porosity modeling based on a facies model using Gaussian of upscaled well logs, 182
Random Function Simulation method, 370 univariate statistics, 195
porosity modeling based on a facies model using Sequential upscaled logs using filters, 184
Gaussian Simulation method, 366 variogram analysis, 201
porosity modeling using secondary data, 360
realizations, 445
sample variograms, 364
R
setting up seed numbers, 104
Random Pick
stochastic techniques, 323
for upscaling discrete logs, 171
truncating distribution range, 328
variogram analysis, 325 Realizations, 445
variogram analysis in gaussian simulation, 362 defined, 433
variogram analysis using correlated attributes, 373
variograms, 335 Regression analysis, 199

Petrophysical modeling using Secondary Data Relative sinuosity, 275

bivariate distribution, 408 Reports
cokriging, 403 volumetrics, 432
collocated co-kriging, 404
local varying mean, 401
Property Modeling Index • 479
S Spatial filters, 45
Spatial trends, 332
Sample variograms, 63, 69
calculation of point data set, 75 Spherical, 57
Saved searches Spherical variogram model, 201
in upscaling, 181
Standard Deviation, 39
Scale up. See Upscaling well logs defined, 26, 40

Scale up well logs, 168 Static volume calculation, 431

defined, 26 Stationarity
Scenario defined, 26, 33
defined, 433 Statistical information, 22
Search Cone, 337 Statistics tab
Search distance, 65 for continuous properties, 197
for discrete properties, 195
Secondary variables, 43
Stochastic simulations
Seed defined, 26
defined, 26 sequential gaussian simulation, 102
setting up seed number, 104
Stochastic techniques
SEED variable in facies modeling, 126
defined, 435 in petrophysical modeling, 130, 323
Semimadogram, 63 Surfaces
defined, 26
Sensitivity, 434
Sequential Gaussian simulation
defined, 102, 115
T
Sequential Gaussian Simulation, 354 Template
porosity modeling based on a facies model, 366 defined, 26
Sequential Gaussian Simulation (SGS) (stochastic), 130 Terminology, 24
Sequential indicator simulation, 127, 245 Toggle
global fraction, 250 defined, 26
how to generate 1D trend input, 251
Train estimation model, 136
how to generate 2D trend input, 252
influence of global facies proportion on facies modeling, 254 Transformation
influence of variogram parameters on a facies model, 253 defined, 26
setup in Petrel, 248
theory, 247 Transformation of data. See Data Transformation
using variograms and trends for facies model generation, 258
Transition lines, 312
with seismic attribute, 261
Trends, 60
Shift scale, 330
defined, 26
Sill, 53 in adaptive channel modeling, 276
in calculating variance, 83 in adaptive chennel modeling, 276
options in Petrel, 332
Simbox mode, 202 using variograms and trends for facies model generation, 258
Simple Kriging, 84 Truncated Gaussian Simulation, 127, 300
Simulation, 101 areas of use, 303
480 • Index Property Modeling
 editing transition lines, 313 Secondary, 43
facies association, 304
Variable spreadsheet
geometry, 311
for uncertainty, 437
global fraction/facies proportions, 304
modeling a transitional depositional environment, 314, 316 Variance, 32, 39, 53
theory, 302 calculating during Kriging, 82
transition lines, 312 defined, 26, 82
variogram and smoothing, 305 making a variance output property, 84
variogram & variance, 311 property, defined, 82
with trends, 306
with trends - principles, 308 Variance Reduction Factor, 406
with trends - process steps, 309 Variogram map, 65
Truncated Gaussian with trends, 127 calculation of a point data set, 73
generating, 66, 70
Truncations
input truncation, 328 Variograms, 231
output truncation, 329 analysis, 201, 362
analysis using correlated attributes, 373
angle tolerance, 64
U applied variogram modeling, 58
bandwidth, 64
Uncertainty behavior, 59
defined, 434 classical, 63
output, 435 creating a variogram map from an acoustic impedance
principle in Petrel, 434 property, 373
realizations, 445 creating a variogram model based on the sample variogram,
results and display, 437 378
Uncertainty analysis, 433 creating sample variogram from an Acoustic Impedance
using the volume calculation case, 441 property, 376
cyclicity, 59
Unconditional simulation, 113 defined, 27, 52
defining a variogram model, 76
Univariate Statistics (Continuous properties), 196
directional variogram analysis for porosity, 347
Univariate Statistics (Discrete properties), 195 exponential, 57
gaussian, 58
Upscaling well logs generating, 231
averaging logs from multiple wells, 179 horizontal variogram maps, 63
cells size in I,J,K direction, 178 influence of parameters, 110
continuous well logs, 172 influence of variogram parameters on a facies model, 253
discrete logs, 170 influences of parameters, 91
how to qc of upscaled logs, 189 in gaussian simulation, 110
how to scale up continuous well logs, 186 in petrophysical modeling, 325, 335
how to scale up discrete well logs, 184 lag, 53
methods, 180 lag tolerance, 65
net/gross, 174 logarithmic, 63
qc logs using filters, 184 model types, 57
quality control, 182 nugget, 53
saved searches, 181 nugget values, 92
well logs, 168 pairwise relative, 63
parameters, 53
V range, 53, 90
sample variogram, 69
Variables sample variograms, 63
primary, 43 search distance, 65
Property Modeling Index • 481
 semimadogram, 63
settings, 72
sill, 53
spherical, 57
trend, 60
types, 63
use of secondary data, 339
using the spherical model to calculate the variance for every
distance, 86
using variograms and trends for facies model generation, 258
variance, 53
variogram map calculation of point data set, 73
Volume calculation, 431
creating a report, 432
using the property model, 439
workflow, 432

W
Water saturation
calculating using permeability/porosity relationship
(J-function) from a macro file, 149
create sw property by using a macro in properties calculator,
160
modeling a SW property using Kriging method, 390
modeling a SW property using kriging method and trend data,
396
Water saturation logs
archie’s equation, 145
creating log by using Archie’s equation, 159
j-function for sw logs, 146
Weighted tab
for upscaling discrete logs, 171
Well correlation
facies modeling, 224
Well Tops
defined, 27

Z
Zonal Anisotropy, 68
Zones
defined, 27

482 • Index Property Modeling