Petrel Workshop

on Property Modeling

Manish Kumar Singh

Abhishek Moharana
Agenda

1. Studio
2. Basic Statistics
3. Geostatistics / Variogram Analysis
4. Geomodeling Basics
5. Data Analysis
6. Property Population Techniques/Approach
7. Property Modeling in Petrel
8. QC grid for Population
9. Case Study
Making knowledge work
Addressing the productivity challenge

What data is available?

Finddone in the past?
What work has been

Collaborate
How do we work together?

Manage
How do we capture and preserve the results?
Multiple Data Sources Find information
In Context at the Enterprise Level
in context of your project
 Collaborate in
context
with your team

Send to Studio
Manage in context
of the team
Take Petrel further with Studio

In-context access to data and information in Petrel, Studio, &

Find
external sources

In-Context
Collaborate
Team collaboration via publish-subscribe model, notifications and
messaging

Management of Petrel data and generated knowledge - models &

Manage
workflows
Basic Statistics
Basic Statistics
Definitions

1. Probability a measurement of the likelihood of an event. (Measured in percent).

2. Variance a measurement of how different the members of a collection are from each
other. (Measured in units of the collection).

3. Correlation a way to measure whether two separate collections are related. (Measured in
percent).

4. Anisotropy a way to measure whether variance within a collection of data is determined

by direction. (Measured in azimuth and percent eccentricity).
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)
Basic Statistics
Histogram and Probability Distribution Function (PDF)

Classes
Number of Samples or
Frequency (%)

PDF

Property Value

Histogram is a graphical representation of the frequency distribution of a

selected variable
Basic Statistics
Cumulative Distribution Function (CDF)

1
Cumulative

CDF
frequency

0
Classes Property Value

The histogram classes are ranked in ascending order and displayed

as a cumulated fraction
Measure of Shape

• Skewness
- coefficient summarizing symmetry
1 N

3
( X i - m)
N 3
i=1
• Coefficient of Variation
- alternative to skewness
 if >1 then check for unusual extreme values
m
Variogram
Analysis
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
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 Do We Need Variograms?

Reasons 1 - Variograms are required for many algorithms

If you choose a Kriging algorithm, or certain other geostatistical algorithms

for modeling, you must have a variogram. In most cases, a default
variogram will be available; making it unnecessary to actually create and
shape the variogram, but still, a variogram is required for many of the
algorithms because the variogram becomes the primary weight function
during modeling.
Why Do We Need Variograms?

Reason 2 - To determine the natural heterogeneity or inherent granularity

of the data in the vertical direction

It turns out that the Range of the Vertical variogram is a good candidate
for the layering increment within a particular zone. We want a layer
thickness that will allow differences in facies to be seen; around one-half
the vertical range is a good start.
Why Do We Need Variograms?

Reason 3 - To determine if there is anisotropy in the horizontal direction

Tools are available during the creation of the horizontal variogram that
allow you to establish if anisotropy exists and to measure it.
Why Do We Need Variograms?

Reason 4 - To have another quality control measurement for comparison

before and after modeling operations

As you analyze, edit, threshold, and model your data, the variogram just
like the histogram is an excellent tool for making sure that the
characteristics of the data are preserved after each modeling operation.
For example, the variogram of the upscaled well logs and the variogram of
the 3D model should be similar, if you are to assume that the proper
algorithm was used
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?
Variogram Parameters

Variance: A measure of how different

members of a collection are from each other.

Variance
Lag distance: Separation distance between Sill
points.

Sill: Variance at the point where the summary

plot flattens out to random similarity. Nugget Range
Separation
Range: Correlation distance; distance 1 2 3 4 5 distance (lag)
beyond which data points no longer exhibit
The Variogram can be calculated in 3
any statistical similarity.
directions:
Nugget: Degree of dissimilarity at zero
distance. • Horizontal Major
• Horizontal Minor
• Vertical
Variogram Calculation
Calculation process and setup: Search radius
1. A Search radius and Lag increment must be Defines how far
defined => Decides the Number of lags and away data will be
consequently the Lag Distance collected
2. All pairs of points in each Lag (bin) will be Lag distance
compared Defines max distance
3. For each Lag (with a given number of pairs), pairs of data should be
the average variation is calculated (squared
difference) apart when compared
(within each lag)

Variogram plot:

Semi variance
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
4. A Regression curve (grey line) is made range
based on all plotted points Lag distance
5. Fit a curve through the Experimental
Variogram to create a ’best fit’ Variogram
Model (blue line) Search readius
Lag distance
Example of Experimental Variogram calculation

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

1 ) 2 1
(  2 - 1 )
))
2
2 2 (3 - 1)
) + 2
3 
( 3 - 2 ) 3 ( 4 -+  2)

2

… + … +
i )
… i …
+ 2
i 1 i 1 - i i 1
) ( i  2 - )i
2
( )
… + i  2 +
… … …

N1 2 N2 2
1 1
 h1 
2 N1
 i 1   i 
i 1
 h 2  
2N2
     
i 1
i 2 i
Result of Experimental Variogram calculation

Nh 2

     
Semi-variogram 1
 h   (i h ) i
can be calculated experimentally as: 2Nh i 1

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?
F =3
F =5 g(h)
F =7
8
Sill
7 g(4)=7.124
F =6 6
g(3)=7.1
F =4 5

F =1 4 g(2)=5.75
F =1
3

2
g(1)=2.214
F =4 1

0
0 1 2 Range
3 4 h5
Variogram Model Types

Variance

Distance

Spherical: Good general algorithm

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

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
 Petrel: Add trend and residual to obtain estimates
There is no apparent Sill

From CV Deutsch, 2002

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
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
High
particle size across vari
abi l i ty
the channels will be acro
Part ss c
much higher than icle han
w i th size nel
along the channels dist
ance
decr
ea
from ses rap
chan id l
nel y
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.
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.

Lag
Y axis

o
0
=6

tole
n

ran
c tio
e

ce
Dir id th

La
dw

g8
B an

La
g7
Angle=60o

La
g6
La
o
5
g5
=2
ce
La

an c e
g4

oler s tan
La

le t d i
g3

An
g Lag
La
g2

La
g1

X axis
Suggested Lag distance: Lateral = well spacing
Vertical = cell thickness
 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
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 icon to measure
3. Click on Run. The result will be placed in the
Input pane or 3D Grid > Variograms folder
the Anisotropy orientation
6. Read the values in the Status bar

1 5

6
Basic Statistics
Variogram Map – Anisotropy

R minorR major

Variogram Map: Sample Variograms :

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

4
 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. Split into both major and minor variogram ranges using the Select and
edit/add point icon 7
8. Open the Variogram Settings to see the Variogram model parameters
(Model type, Sill, Nugget, Orientation and Ranges)

5
6
Sill

Nugget Range 8
 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
Kriging
Influence of the Variogram model parameters

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

Exponential model Spherical model Gaussian model

Kriging
Influence of the Variogram model parameters

Variogram Range (spherical variogram)

Range: 1000m Range: 10000m

Kriging
Influence of the Variogram model parameters

Azimuth (spherical variogram)

Bad Good

Anisotropic: +45 degree Anisotropic: -45 degree

Range: 10000m / 5000m Range: 10000m / 5000m
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
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
Basics Rules of Modeling
Modeling
Need to predict the behavior of a reservoir
property at unsampled locations based upon
sampled data

?
?
? ?
?
?
 Dilemma
Very few direct observations - many indirect
• Permeability measurements from core analysis
• 10’s of well test surveys
• 100’s of wells with diverse suites of logs
• 1,000’s-1,000,000’s of seismic traces

Indirect measurements are of lesser quality

• Well test yield estimates averaged over large volumes
• Well logs may yield only good estimates of porosity
• Seismic data may yield only good estimates of impedance
with poor vertical resolution
Concerns

How is the integration performed?

• Varying quality of the various data types
• Data are not of the same variable
• Data are not the of the same domain (time/depth)
• Some data are 2D maps

How good is the resulting reservoir model?

• Uniqueness of the model
• Uncertainty associated with estimated properties
• Appropriateness of fluid-flow simulation model
• Impact on economic analysis
Data Types Availability
1. Well Log Data (surface tops, rock type, φ, Κ) by zone
2. Core Data (φ and Κ by rock type) by zone
3. Sequence Stratigraphic Interpretation / Layering (a definition of the continuity and trends
within each layer of the reservoir)
4. Trends available from a regional geological interpretation
5. Analog data from outcrops or densely drilled similar fields (size distributions, measures of
lateral continuity)
6. Seismic-Derived Attributes
7. Well Test and Production Data

This information is sparse relative to the size of the heterogeneities being modeled; therefore,
there is always uncertainty in the geological model
Misconceptions about Geostatistics

1. Does not create data

2. Does not replace the need for data
3. Does not replace geologic expertise
4. “will not replace Geoscientists”
5. Does not make decisions
6. Not solely domain of statistician
Criteria for Model Verification
Things that should be checked:
1. Does the model appear geologically plausible?
2. Relative to other models, are the heterogeneities reasonably distributed?
3. Are the porosity and permeability models consistent with the rock type model?
4. Does the model present trends consistent with the regional geology?
5. Does the model honor all of the input data?
6. Variograms modeled correctly
7. Can the model be checked relative to data that were not used in the model
building, e.g., well test or production history?
Data Analysis
Data analysis

Data Analysis is a process of quality controlling the data,

exploring the data and preparing inputs for Facies and
Petrophysical modeling.
Why do Data Analysis?

Is there a pattern by which the facies proportion changes

laterally and vertically? Is there a trend in the data set?

Is there any pattern of correlation between a seismic

attribute (say AI) and the facies?

What are the variogram parameters like Range (in the

major, minor and vertical directions), nugget, and the
variogram type etc. for the properties to be modeled.
Basic Statistics
PDF and CDF in Petrel (Histogram) Histogram in object settings

Histogram & CDF in a Histogram window 1. Open the object Settings window
2. Go to the Histogram tab
1. Open a Histogram window 3. Use filters and intervals/increments as
2. Select the property to plot needed
3. Select the Show cdf curve icon
4. Use filters if needed
Basic Statistics
Normal Score Transformation
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 has insufficient data Linear regression and
(e.g. only a few wells) correlation coefficient
Basic Statistics
Generic Filters in Petrel (Histograms and Crossplot)
Different filters can be applied in Petrel, but Generic filters from 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
General settings for Data Analysis
What data are to be used in Data Analysis?

Two types
Discrete data.
Channel sands Code 0
Levee Code 1
Crevasse Splay Code 2
Flood plane Code 3

Continuous Data.
Porosity
Permeability
water saturation
Types

Discrete property, you can analyze the facies proportion, facies

thickness, calibration between continuous properties (e.g.
sampled seismic) and facies within each zone and also create a
variogram.

Continuous property, you can define data transformations and

generate variograms
Which variable to analyze first?

Aim is to define the conceptual facies model first and

then model the petrophysical properties according to
the facies model

1. Data Analysis of Facies

2. Modeling of Facies
3. Data Analysis of Petrophysical properties
4. Modeling of Petrophysical properties by conditioning to facies
How to do Data Analysis on Discrete variables?

1. Vertical Proportion curve

How the proportion of each facies varies vertically in a

stratigraphic zone.

if the environment being modeled, has a progressive downward

increase in shale or progressive downward increase of good
porous limestone in a zone, the vertical proportion curve also
should reflect this.
How to do Data Analysis on Discrete variables?

2. Attribute Probability Curve

If the attribute is only present at the well locations this option

should not be used

3. Indicator Variograms
Variograms calculated for discrete data are called Indicator
Variogram
How to do Data Analysis on continuous variables?

Data Analysis on continuous properties includes:

1. Viewing the data distribution as histograms etc.

2. Identifying trends within the data.

3. Making data transformations.

4. Describing the data’s spatial variation by generating variograms.

Requirements of the Modeling algorithms in Petrel to
model continuous properties?
Data is stationary and the mean does not change with direction.
Ie, the data does not contain any trends.

Certain algorithms (simulation algorithms) have a mathematical

requirement of data following a Normal distribution with mean of
zero and standard deviation of 1.
Why do Data Transformation?

To be precise, the data transformations are done to

1. Remove any trends or Non stationarity in the data

2. Convert the data from its original distribution to a Normal

Distribution (with a mean of 0 and standard deviation of 1)
What if there is a trend in the data?
How to remove it ?
How to add it back??

Remove this non stationarity

certain data transformation like 1D

trend or 2D trend or 3D trend has
to be used.

(1D trend is for log data, 2d trend s

for data in 2 dimension like
surfaces, scatter points , and 3d
trends are used for 3D properties).
What is Normal Score Transform (NST) or Gaussian
Anamorphosis

Normal distribution :Mean of Zero and Standard Deviation of 1

Normal Score Transform (NST) or Gaussian Anamorphosis

Process by which the distribution of the input data is

changed to a Normal score domain (the distribution now
becomes a bell shaped curve with mean=0 and standard
deviation=1)

Deterministic approach like Kriging this is not required.

How to do back transformation?

1. Automatically by the Programme.

2. End product after modeling is a data set that follows the

original distribution and that preserves the original trends
How to do variography of continuous variables?

1. Without conditioning to facies

2. By conditioning to facies
Variation of PHI within a channel will have a different pattern as
compared to the pattern of PHI variation in the background flood

plain
How to do variography in Horizontal directions.

1. Use conceptual knowledge of the area.

2. Using a seismic attribute as a Proxy for horizontal

Variograms
For example, the model data contains Acoustic Impedance and

Porosity data which are showing a good correlation

How different parameters affect the Variograms and
hence the results?

Range:
simple words,
1. Large range means greater continuity
2. Small range means less continuity
3. The larger the range the smaller the heterogeneity
Property Population
Approach
Steps in Constructing 3-D Models
1. Determine the areal and vertical extent of the model and the geological modeling cell size
2. Establish a conceptual geological model and define zones for modeling
3. For each zone:
(a) Define stratigraphic correlation
(b) Define the number of rock types, the rock type data, and their spatial correlation
(c) Generate 3-D rock type model
(d) Establish the porosity and permeability values from core / log data and the spatial correlation
(e) Generate 3-D porosity models
(f) Generate 3-D permeability models
4. Verify the model
5. Uncertainty is assessed by deriving reasonable estimates of uncertainty for each input parameter and
then generating multiple realizations
Deterministic Approach

Sufficient information of the process is

known, so that an accurate model may be
determined from a few samples

Deterministic modeling leads to a single

unique solution
Stochastic Approach

Sparse information known

Performs the integration of the data types

providing information at different scales and
with different levels of certainty

Challenge is to choose the best outcome

relevant to final goal of the reservoir model
Property Population Fundamentals

Geostatistical Algorithms
 Ordinary Kriging
 Simple Kriging
 Cokriging
 Collocated Cokriging
 Sequential Gaussian Simulation
 Indicator Simulation
Needs a Variogram Model
Estimation
(Deterministic)
Krigging
Kriging is a geostatistical estimation technique that uses a linear
combination of the sample data to make estimates of the data at an
unsampled location.

Kriging is often associated with the acronym B.LU.E. for “best

linear unbiased estimator”.

• Best: Minimizes the variance of the estimation errors.

• Linear: based on a linear combination of the weights applied

to the data samples.

• Unbiased: tries to make the mean residual equal to zero.

Kriging Theory

Kriging uses the variogram to understand the variability of

the data over a distance.
This knowledge allows kriging to calculate weights that
minimizethe error variance.
It also accounts for other factors that could affect the
weights:
 closeness to the location being estimated
 redundancy between the data values
 anisotropic continuity (preferential direction)
 magnitude of continuity / variability
 Cokrigging
Cokrigging needs both primary and secondary input
 Output is a weighted combination of primary and
secondary data
 Both primary and secondary data can be scattered
 Requires variograms for both primary and secondary data
coupled with cross-variograms
 Correlation should be reasonable between primary and
secondary data
 Very CPU intensive
 Collocated Cokrigging

Collocated cokriging needs both primary and secondary input

 Secondary data should be present at all output locations of
primary variable
 Linear trend should exist between primary and secondary
data
 Requires only the primary variogram
 Less CPU intensive than Cokrigging
Simulation
(Stochastic)
Analogy Stochastic Modeling
“Its is better to have uncertainty
rather than illusion of reality”
-Andre G. Journel
 Sequential Gaussian Simulation
Sequential Gaussian simulation is used for continuous
variables
 Needs to have data in Normal Score Transform
 Output is more realistic owing to heterogeneity

Procedure -
- Obtain a random path
- Krige the Value at Unsample Location
- Obtain the local ‘cdf’ from local mean and SD.
- obtain another random path and repeat the process
to generate several realizations
SGS Example
Indicator Simulation
Indicator Simulation is used for discrete variable
such as litho types, rock class, facies

• input data should be classified in indicators or categories

(integer values)
• krigging is used to estimate the indicators
• generated grid can be used as a template grid to guide
other property or to selectively populate properties for
specific litho types
• similar to SGS.
Indicator Simulation Example
Porosity Modeling
Example
Porosity Modeling Scenarios

Depends on data quality and availability

• Well Logs only
• Well Logs + Seismic
• Well Logs + Seismic + Lithology
• Stochastic Modeling (multiple realization)
Wells Logs Only

Data Input
• Porosity Logs
• Variogram Model of Porosity (Geostatistical Methods only)
• Population Parameter
Well Logs Only - Ordinary Krigging Result

Smooth Output
with Trend (Areal)
Well Logs + Seismic

Data Input
• Porosity Logs
• Seismic Attribute
• Correlation Coefficient
• Porosity Variogram
• Population Parameter
 Well Logs + Seismic - CoK Result

Away from the wells

Porosity is guided
by Seismic
Well Log + Seismic + Lithology

Data Input
• Porosity Logs
• Seismic Attribute
• Correlation Coefficient
• Lithology Model
• Porosity Variogram
• Population Parameter
Well Log + Seismic + Lithology - Result

Heterogeneous output
incorporating lithology
Stochastic Modeling

Data Input
• Porosity Logs
• Data Transform
• Porosity Variogram
• No. of Realizations
• Population Parameter
Petrophysical Modeling using Secondary Data

 Local Varying Mean:This will perform a normal

transformation on both the secondary property and the
upscaled cells before calculating the residual in normal
score space

 Collocated Co-Kriging: This will use a correlation coefficient

together with the secondary variable to calculate the
contribution of the secondary variable at each point
Kriging Vs. Simulations

Kriging: SGS:
Is the most probable value for each cell.
• Is a possible value for each cell.
Doesn’t respect spatial variability far from
data • honors spatial variability
Don’t respect data histogram • Respects data histogram
Smooth: • appropriate for flow simulation where
 appropriate for visualizing trends, extreme values are important,
• Does assess global uncertainty
 inappropriate for flow simulation where
extreme values are important, • Several equiprobable simulations
Does not assess global uncertainty
One unique kriging
Selecting an Algorithm for Petrophysical Modeling
Property Modeling in Petrel
Property Modeling
Data Integration
Input Data
Well data, seismic, production, previous geological studies
Reservoir Modeling >> Geostatistics

Geology Geophysics Petrophysics Engineering

Deterministic
Information
(framework)
Integrated Study
Conceptual Statistical
Integrate maximum amount of Information
Information
information (connectivity) (variation)
Petrel Workflow Tools
Introduction Surfaces and Stratigraphic
Data edit Modeling 3D Grid Construction: Structural Modeling
Intro to Petrel
Interface
Studio

Pillar Make Horizons

Fault Modeling Gridding Zones & Layering
3D Grid Construction 3D Grid Construction Property Modeling Make contacts & Volume
Structural Framework Structural Gridding Calculation

Workflow Editor
Facies &
Petrophysical
Modeling
Well Log Upscale
 Processes used for 3D Grid construction

3D Grid Construction

Corner Point Gridding Structural framework

Structural gridding Structural modeling

Fault Framework Modeling Fault Modeling Fault Modeling Framework

Fault model from Structural
framework
Structural Gridding Pillar Gridding Horizon Modeling

Make Horizons/Zones/Layers Make Simple Grid

Property Modeling
Make Horizons Process

This process will define the vertical

layering of the 3D grid. The layering can
later be refined using the Make
Zones/Layering process.

Structural modeling

Fault modeling Pillar Gridding Make Horizons Make Zones Layering

Make Zones - Results

Before After
Layering - Results

Types of zone division:

Follow base

Proportional

Follow top

Fractions

Follow top + using a

depositional surface

Note: It is always a good idea to display

edges in 3D to see all updates before
applying the Make Zones and Layering
Processes.
Property modeling overview
Property modeling is the process of filling the cells of the grid with discrete (facies) or continuous
(petrophysics) properties.. Property modeling is split into three separate processes:

Geometrical Modeling Facies Modeling Petrophysical Modeling

Geometrical Modeling
Geometrical modeling is the process where properties
can be generated by using pre-defined system
Geometrical modeling is the process where properties can be generated by using pre-
variables. No interpolation of input data is required and
defined system variables. For most properties no interpolation of input data is required
properties are built based on the geometrical properties
and properties are built based on the cell geometry. For some though, input data is
of the grid cells themselves, such as a cell volume,
required e.g. Seismic resampling and Geobody modeling.
angle, height, etc. Each cell will get a numerical value
corresponding to the selected system variable. These
properties can be important in processes such as
volume calculations and mathematical
Facies Modeling
Geometrical modeling is the process where properties
canFacies modelingbyis using
be generated a means of distributing
pre-defined systemdiscrete data (e.g. facies) throughout the
3D model.
variables. No The modelingofinputs
interpolation input include
data is the upscaled
required and facies well logs and modeling
parameters.
properties are built based on the geometrical properties
of the grid cells themselves, such as a cell volume,
angle, height, etc. Each cell will get a numerical value
corresponding to the selected system variable. These
properties can be importantFacies Modeling process
in processes such as
volume calculations anddialog
mathematical

Upscaled well
logs
Petrophysical Modeling
Petrophysical modeling is the interpolation or simualtion of continuous data (e.g. porosity,
permeability ) throughout the model grid. Input including well data, facies realization,
variograms, a secondary variable and/or trend data can be used as input and various
user settings are available.

Upscaled well
logs
Trend

Facies
realizations
Variogra
m
Stochastic vs. Deterministic modeling methods

Stochastic Deterministic
Random (Seed number) It is unlikely due to unpredictable factors

It generates different equiprobable results It generates the same result for a given set
for different seed numbers of initial conditions

Variable states are described by probability Variable states are described by unique
distributions values

It does not need upscaled cells: Need upscaled cells – needs more data
Unconditional modeling
Allows more complexity and variability in Faster to run
the model – can help assess uncertainty
Facies modeling methods - overview

Deterministic Learning system

Estimation Direct Addressing Artificial

Indicator Kriging Asign values Interactive Neural Net

Discrete distribution of Choose from Allows the user to Uses the classification
the property honoring undefined, constant, paint facies directly on model made in the Train
the pre-defined other property, surface the 3D model. Estimation Model.
histogram and vertical function.
Facies modeling methods - overview

Deterministic Learning system

Estimation Direct Addressing Artificial

Indicator Kriging Asign values Interactive Neural Net

Facies modeling methods - overview
Stochastic
Pixel based Object based

Sequential Truncated Truncated Multi-point Object Modeling

Indicator Gaussian Gaussian Facies
Simulation Simulation Simulation with Simulation
trends
Distributes the Used mostly Distributes the The variogram is Allows to
property, using with carbonates facies based on replaced by a populate a
the histogram. where facies a transition training image discrete facies
Directional are known to be between facies giving both the model with
settings, such as sequential, it and trend facies and the different bodies
variogram and deals with large direction. The relative position of various
extensional amounts of input trends are to each other, geometries,
trends, are also data, such as converted into describing the facies and
honored. global fractions probabilities to spatial fraction
and trends. then run TGS. correlation from
one-to-multiple
points.
Facies modeling methods - overview

Stochastic
Pixel based Object based
Sequential Truncated Truncated Multi-point Facies Object Modeling
Indicator Gaussian Gaussian Simulation
Simulation Simulation Simulation with
trends
Petrophysical modeling methods - overview

Deterministic
Estimation Interpolation
Kriging Kriging Kriging by Closest Functional Moving
Interpolation GSLIB average
Honors well It performs It has the It uses the I honors well Based on the
data, input fastest. It has a option of closest well and trend data input it gives
distributions, co-kriging collocated co- data input for creating a 3D an average
variograms and option and kriging and you each function value and
trends. It can allows user to can choose unsampled (parabolic, calculates the
work in real choose between location. simple weigths
coordinates between ordinary or parabolic, according to
and it’s fast. simple and simple kriging. planar or bi- the distance
ordinary linear) used in from wells.
kriging. the
interpolation.
Petrophysical modeling methods - overview

Deterministic
Estimation Interpolation
Kriging Kriging Kriging by Closest Functional Moving
Interpolation GSLIB average
Petrophysical modeling methods - overview

Deterministic Learning system Stochastic

Direct Addressing Artificial Pixel based
Assign values Neural Net Sequential Gaussian Gaussian Random
Simulation Function Simulation
Choose from Uses the Honors well data, input It is faster than SGS,
undefined, constant, classification distributions, and gives better
other property, model made in variograms and trends. variogram
surface and vertical the Train The variogram and reproduction.and has
function. Estimation distribution are used to a fast collocated co-
Model. create local variations, simulation option.
It can be used even away from input
data.
Petrophysical modeling methods - overview

Deterministic Learning system Stochastic

Direct Addressing Artificial Pixel based
Assign values Neural Net Sequential Gaussian Gaussian Random
Simulation Function Simulation
QC Grid for Population
QC during Workflow
1. Analyze/understand your data with various geostatistical tools
– Univariate statistics
– Histogram
– Variograms
– Etc...
2. Transform input data as required (threshold, lump, upscale ...) for your data

3. Perform quality control checks to make sure that the transformed data has the
same characteristics as the original data.

4. Use the selection criteria/choose an algorithm appropriate for the data

5. Execute the algorithm to populate the grid

6. Perform quality control checks

QC during Workflow

 Quality Control with Histograms

Perform the same comparisons as above, but use the histogram for
comparison. Here, you want to make sure that the condensed data has
the same basic shape as the original data, and that the histogram of the
output grid has the same basic shape of the condensed data.

 Quality Control with Variograms

Perform the same comparisons as above, but use variograms as the
criteria. Variogram shapes and other characteristics should remain the
same to verify that the character of the output grid reflects that of the input
data.
3D Grid Quality Control Tools
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

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.
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 for Lag spacing =

vertical analysis
1-2 x the sampling
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)
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!

Quality Control of Structural Model 3D Grid Cells
Geometrical properties considered for this analysis are usually the
following:
 Cell volume: Calculates the bulk volume of each cell in the 3D grid.
Shows existence of cells with negative volume.
 Cell height: Calculates the height of each cell in the grid. Finds cells
with minimum thickness or thickness very close to zero that may bring
convergence problems in Simulation.
 Cell inside out: Gives information about very distorted cells inserted
inside other cells or inserted inside themselves.
 Cell angle: Measures deviation between angles of each cell when
compared with a 90o angle. Shows presence of cells with very acute
angles.
Quality Control of Structural Model 3D Grid Cells

1. Bulk volume: Values must be positive

2. Cell height: Minimum value should not be very close to cero

or relatively insignificant

3. Cell inside out: Values must be cero

4. Cell angle: Values should be approximately equal or less

than 60, though higher values are not critical
Statistics
Statistics
Thank You
 Porosity Modeling
Using Acoustic Impedance Cube as Secondary
Information
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

Þ If no density log is available use the sonic log instead of the impedance log for the
crossplot
Þ If this test fails the AI cube cannot be used as secondary input for the 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.
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!

 Case Study
Influence of Layer thickness
Upscaled Imp log – Resampled AI relationship

Layer Thickness:
5m
3m
1m
Impedance Log
Upscaled

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
Upscaled Porosity Log

Resampled Seismic AI

Smaller layer thickness reduces correlation

Case Study
Seismic AI and Porosity Relationship by Zone

Top reservoir Bottom reservoir

Bad correlation
Ok correlation
Upscaled Porosity Log

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
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)