Sie sind auf Seite 1von 213

Reservoir Modeling with GSLIB

Overview

• Objectives of the Course


• What is Geostatistics?
• Why Geostatistics / 3-D Modeling?
• Uncertainty Quantification and Decision Making
• Heterogeneous Reservoir Modeling
• An Example / Limitations / Future

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Objectives of the Workshop
• Increase awareness of geostatistics and the situations where the application of
geostatistical techniques could add value
• Learn how to assemble the required data for geostatistical techniques
• Introduce statistical and geostatistical techniques available in GSLIB
• Provide some statistical / geostatistical tools that will help with specific tasks
(core - log data, permeability prediction, 3-D model building, and uncertainty
assessment)
• Step through major components of a reservoir modeling study (layering, rock
type modeling, porosity modeling, permeability modeling)
• Understand the limitations of the resulting numerical models and the
geostatistical techniques
• Note: while the context and examples in this workshop are specific to the
petroleum field, these tools and techniques can be applied in similar manner to
other areas which encounter geostatistical problems (i.e. mining)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Historical Perspective
• Theory of probability (in its modern form) was formalized in the 1600's by
Blaise Pascal and Pierre de Fermat (Gauss and Bayes were more recent
players)
• The foundation for geostatistical techniques was established by people like
Kolmogorov, Weiner, Matern, and Gandin in the early 1900's
• Geostatistics was started in the 1960's by Krige and Sichel in South Africa and
Matheron in France. Two of Matheron's first students (Journel and David)
would leave for the USA and Canada and start new centers of geostatistical
research
• The application of geostatistical techniques became popular in the mining
industry and meteorology. Now, these techniques are applied in many diverse
applications from fisheries, forestry, environmental remediation, and so on
• Extensively used by major oil companies
• Centers for research are numerous, including Stanford, Fountainbleau, and
others such as the University of Alberta

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Geostatistics for Reservoir
Characterization (1)
• Business Need: make the best possible reservoir management
decisions in the face of uncertainty. One of the biggest uncertainties
is the numerical description of the reservoir.
• 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.
• Geostatistics is a branch of applied statistics that places emphasis on
(1) the geological context of the data, (2) the spatial relationship
between the data, and (3) data measured with different volumetric
support and precision.
• Geostatistics is sometimes referred to as stochastic modeling,
geostatistical reservoir characterization, conditional simulation
...
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Geostatistics for Reservoir
Characterization (2)
• Basic Principles:
– work within all known geological (physical) constraints
(sequence stratigraphic framework, …)
– provide tools to quantify and exploit spatial correlation
– algorithms for numerical geological modeling
(heterogeneity modeling) and uncertainty quantification
• Doesn't make reservoir modeling any easier; just
better (if correctly applied)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Motivation for Reservoir Models
• There is a need for reliable estimates of the original volume of hydrocarbon in
the reservoir. These insitu volumes are important to (1) determine the
economic viability of producing a given reservoir, (2) allocate equity among
multiple owners, (3) compare the relative economic merits of alternative
reservoirs, and (4) to determine the appropriate size of production facilities.
• Well locations must be selected to be economically optimal and robust with
respect to uncertainty in the reservoir description. What type of wells
(horizontal, vertical, multilateral, ...? How many wells?
• Reconcile an abundance of soft data (say, from a 3-D seismic survey or
historical production data) with a limited amount of hard well data.
• Assess the potential for bypassed oil and the value of infill wells
• Flow simulation predicts reservoir performance with different production
scenarios. The use of flow simulation was initially hampered by the limited
resolution of flow models (primarily a computer hardware consideration) and
simplistic geological input.
• Modern decision analysis tools require an assessment of the uncertainty in
future production. One of the greatest sources of uncertainty is uncertainty is
the geological inputs.
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Key Geostatistical Concepts (1)
• Petrophysical Properties: concerned with constructing high resolution 3-D
models of lithofacies types, porosity, and permeability.
• Hard truth measurements are the lithofacies assignments, porosity,and
permeability observations taken from core (perhaps log) measurements. All
other data types including well logs and seismic are called soft data and must be
calibrated to the hard data.
• Modeling Scale: It is not possible nor optimal to model the reservoir properties
at the resolution of the hard core data. The core data must be scaled to some
intermediate resolution, models are generated at that intermediate geological
modeling scale, and then the geological model is possibly scaled to an even
coarser resolution for flow simulation

~2 cm ~10’s m ~100’s m
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Key Geostatistical Concepts (2)
• Numerical Modeling: At any instance in geological time, there is a single true
distribution of petrophysical properties in each reservoir. This true distribution
is the result of a complex succession of physical, chemical, and biological
processes. Although some of these depositional and diagenetic processes may
be understood quite well, we do not completely understand all of the processes
and have no access to the initial and boundary conditions in sufficient detail to
provide the unique true distribution.
Point bar sand
Braided channel sand
Crevasse sand
Coal
Mudfilled channel
Floodplain shale

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Key Geostatistical Concepts (3)
• Uncertainty: All numerical models would be found in error if we were to excavate that
interwell volume and take exhaustive measurements: There is uncertainty. This
uncertainty exists because of our ignorance/lack of knowledge. It is not an inherent
feature of the reservoir.
• Uniqueness and Smoothing: Conventional mapping algorithms were devised to create
smooth maps to reveal large scale geologic trends; for fluid flow problems, however,
the extreme high and low values often have a large affect on the flow response.
• Analogue Data: There are rarely enough data to provide reliable statistics, especially
horizontal measures of continuity. For this reason, data from analogue outcrops and
similar more densely drilled reservoirs are used to help infer spatial statistics that are
impossible to calculate from accessible subsurface reservoir data.
• Dynamic Reservoir Changes: Geostatistical modeling provides static descriptions of
petrophysical properties. Time dependent changes in pressure and fluid saturations are
best modeled with a flow simulator that encodes physical laws such as the conservation
of mass and energy.

Flow
Simulation
Distribution of
Multiple stochastic models possible responses

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Cell-Based Modeling
Lithology

Porosity

Permeability

• Cell-based, stochastic, 3-D modeling of lithology, porosity and


permeability in a sequential order
• Preserves major heterogeneities and statistical features
• Statistical control is obtained from cores, logs, seismic and outcrops

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Object-Based Modeling

• Object-based, stochastic, 3-D modeling of well defined


geometric objects
• Pseudo genetically simulate depositional history
• Statistical control is obtained from cores, logs, seismic and
outcrops
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Decision Making in the Face
of Uncertainty

Given 49 balls numbered consecutively from 1 through 49, what is the


average of the numbers showing on six balls drawn at random from the
49?
• The honest answer is somewhere from 3.5 to 46.5
(13,983,816 combinations)

Frequency
• The most likely value is 25
• The optimal answer depends on two things:
1. the uncertainty in the average, and
2. the impact of making a mistake
• Stochastic simulation quantifies uncertainty⇒
3.5 25.0 46.5
in a distribution of uncertainty
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Impact of Making a Mistake:
Loss Functions $
An Example Decision:
• Do we carry an umbrella?
– Uncertainty: will it rain?
– Impact of an underestimate: get wet
– Impact of an overestimate: carry it around for nothing
• How tall should we design a dam?
– Uncertainty: what is the largest rainfall in the lifetime of the dam?
– Impact of an underestimate: failure of dam and loss of property
– Impact of an overestimate: additional material and labor cost
• Do we clean a potentially contaminated site?
– Uncertainty: what is the contaminant level?
– Impact of an underestimate: insurance claims/lawsuits due to
health problems
– Impact of an overestimate: cost of unnecessary cleaning
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Optimal Estimates
(depend on the uncertainty and the impact of an mistake)
Loss function quantifies the impact of an mistake:

Cost $ Cost $ Cost $

Underestimation Overestimation Underestimation Overestimation Underestimation Overestimation

The distribution of uncertainty quantifies the uncertainty:


Frequency

Frequency

Frequency
Optimal Estimates
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Uncertainty
(decreases as more data becomes available)
Returning to the previous example, what would be the average of six balls drawn
from 49 balls, numbered 1 to 49?
Knowing the outcome of Knowing the outcome Knowing the outcome
one ball: of two balls: of three balls:

43 43 19 43 19 23
Frequency

Frequency

Frequency
Note the decrease in uncertainty
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
An Application to Miscible
Flood Design
Srivastava, 1990
The decision to be optimized:
How much solvent should be injected?

• The answer depends on the connected pore volume.


• The uncertainty in the connected pore volume can be quantified with
geostatistical simulation.
• The loss function is a function of the cost of solvent and the oil price.

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Geostatistical Simulation
Multiple equiprobable realizations of the reservoir

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Uncertainty in Connected Pore
Volume
Running a random-walk sweep program on each geostatistical
simulation realization provides a distribution of the connected
pore volume: Connected Pore Volume
Frequency

Volume
Now, apply loss function and compute how much solvent to inject

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Decision Making in the Face
of Uncertainty
Loss Function Distribution of Uncertainty and
Optimal Decisions
|slope|=a=$oil-$solvent
$

|slope|=b=$solvent

Underestimation Overestimation

The optimal decision are indicated for various


oil to solvent cost ratios.

Single answer integrates uncertainty and the impact


of a mistake
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Decision Making in the Face of
Uncertainty
• Stochastic simulation is used to generate plausible
realizations of what reality might be like
• Stochastic simulation is also used to quantify
uncertainty
• Loss functions are appropriate for quantifying the
impact of making a mistake
• Together, simulation and decision analysis tools,
are appropriate for reservoir management
• This workshop will focus on stochastic simulation
as applied to create plausible reservoir models
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Stochastic Reservoir Modeling
A comparison between reality and a numerical model
Reality Model
Distribution of Rock/Fluid Distribution of the Rock/Fluid
Properties Properties

multiple stochastic models


single true distribution
Recovery Process Recovery Process

numerical model
actual process implemented of process
Field Response Single true response Field Response Distribution of
possible responses

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Constructing 3-D Models
The specific process employed for 3-D model building will depend on the data available, the
time available, the type of reservoir, and the skills of the people available. In general, the
following major steps are required:
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 data, and the spatial correlation
c) Generate 3-D rock type model
d) Establish porosity and permeability values and the spatial correlation
e) Generate 3-D porosity models
f) Generate 3-D permeability models
g) Merge and translate back to real coordinates
4. Verify the model
5. Combine zones into a single model
Each of these steps is addressed to some extent in this workshop

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Data for Reservoir Models
Core Production
Log Conceptual Data
Reservoir Model
Seismic

• Core data (φ and Κ by lithofacies)


• Well log data (stratigraphic surfaces, faults, measurements of petrophysical
properties)
• Seismic-derived structural data (surface grids / faults)
• Seismic-derived attributes (vertically averaged facies proportions and φ)
• Well test and production data (interpreted K •thickness,channel widths,
connected flow paths, barriers)
• Sequence stratigraphic interpretation / layering (a definition of the continuity
and trends within each layer of the reservoir)
• Spatial patterns from regional geological interpretation
• Analogue data from outcrops or densely drilled similar fields (size distributions,
measures of lateral continuity)
• Knowledge of geological processes / principles established through widely
accepted theories (forward geologic modeling)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Recall of Geological Gridding (1)
• Triangular facets allow flexible modeling of surface
grids and have some significant advantages:
– handle multiple Ζ-valued surfaces
– natural surface gridding for tetrahydra-based volume gridding
• In general, however, we prefer regular nx •ny Cartesian
grids:
– easier to work with cells that have the same volume
– tetrahydra topology not commonly accepted by finite-
difference scale-up and flow simulation programs
– most visualization and mapping programs have been designed
for either corner-point or block-centered Cartesian grids.
...
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Recall of Geological Gridding (2)
•Also prefer a vertical Z coordinate, i.e.,

Two approaches to establish areal gridding: (1) on the left, the cells/nodes are
stacked vertically, and (2) on the right, the cells/nodes are aligned on lines
perpendicular to the bounding surfaces. The first approach, with a universal
vertical coordinate, is preferred because of computational simplicity.

• Work within “stratigraphic layers” defined on the basis of:


– sequence stratigraphic zonation
– keep geologically “homogeneous” rock together
– maintain a reasonable number of data per zone
– less resolution outside the volume of interest

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Stratigraphic Layers for
Modeling(1)
• One example using top surface as datum (total vertical extent about 200
meters, horizontal extent about 15 kilometers):

Surface Type:

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Stratigraphic Layers for
Modeling(2)
• Another example for a deep water depositional system:
Surface 5 100m elevation
Surface 4
Surface 3
Surface 2

Surface 1 0m elevation

• Another example:

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Stratigraphic Coordinates in
Presence of a Fault
Stratigraphic Space

Grid Model
Pre-disturbance
Stratigraphic Space
Post-disturbance

Grid translations may be applied to “remove” faults and tilting in order to


work in the predisturbance stratigraphic setting.

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Advantages of Geostatistics
• Intellectual integrity (?) Mathematical consistency (?)
• 3-D models lead to better volumetrics
• Better modeling of heterogeneity
– no need for pseudo wells
– controllable degree of spatial variability
– flow models are more reliable
• Framework to integrate data
– geological interpretation
– core and log data
– seismic data
– production data
• Assessment of uncertainty in process performance due to uncertainty in
geological model

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Reservoir Modeling with GSLIB

Introduction to GSLIB

• Answers to Some Commonly Asked Questions:


– What is GSLIB and how do I get it?
– How do I get it to work?
– What are the strengths and weaknesses of GSLIB?
– What is the future of GSLIB?
– What about user support?
– Are there any restrictions on its use?
• Alternatives and Supplements to GSLIB
• Geostatistical Reservoir Modeling
• GSLIB Preliminaries

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
What is GSLIB?
• GSLIB is an acronym for Geostatistical Software LIBrary. This name was
originally used for a collection of geostatistical programs developed by
students and faculty at Stanford University over the last 15 years.
• The original GSLIB inspired the writing of GSLIB: Geostatistical Software
Library and User's Guide by Clayton Deutsch and André Journel during 1990
- 1992. This publication (available for about $50 from Oxford University
Press) contains a CD and a user's guide for code based on the original GSLIB.
• Executables and source code available at www.gslib.com
• Most of the original GSLIB code was rewritten for uniformity and new
programs were added during development
• A second edition was completed in March, 1997
• Executables available at www.gslib.com
• A third edition is planned for 2002

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
How do I get it to work?
• There are 39 FORTRAN programs on the GSLIB CD-ROM
• Download executables for PC
• You should have a FORTRAN compiler
• You may need to edit some of the included files to increase the allowable
number of data and grid sizes (FORTRAN 90 version overcomes this problem)
• Data files are ASCII flat files
• Parameter files are ASCII files (see example on page 12 of this lecture)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
What are the strengths and
weaknesses of GSLIB?
Strengths:
• one of the most comprehensive geostatistical packages available
• source code is provided
• flexible and portable
Weaknesses:
• no user support
• the user interface is not “friendly”
• limited object-based modeling capability
Toolbox and not an integrated package

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
What is the future of GSLIB?
second edition is just finished -- bug fixes and minor changes for a few years

What about user support?


there is none

Are there any restrictions on the


use of GSLIB?
acknowledge where it came from

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Alternatives and Supplements to
GSLIB
• There are commercial modeling packages
• There are some GSLIB add-ons and GUIs developed at different places ...
• A good statistical analysis package would be a useful supplement
• A contouring / mapping package would be useful
• There are some additional SCRF software such as variogram modeling and 3-
D model display that would be useful

Warning / Disclaimer
• Reservoir modeling requires a sound geological understanding of the reservoir,
an appreciation for the quality of the available data, an understanding of the
objectives of the modeling exercise, and so on. This course does not teach
reservoir modeling; the goal is to make you more familiar with the tools in
GSLIB that may be helpful in a reservoir modeling / geostatistical study.

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Programs in GSLIB (1)
• PostScript plotting utilities (histograms, probability plots, Q-Q/P-P plots,
scatterplots, location maps, gray/color scale maps)
• Utility Programs:
– cell declustering
– data transformation
– IK manipulation: order relations, change of support, p or q values, E-type
estimate,
– other theoretical indicator variograms from multiGaussian distribution,
linear system solvers, data calibration (Markov-Bayes), ...
• Measures of spatial continuity (Variograms):
– direct and cross variograms of any # of variables
– 1,2, or 3-D regular or irregular spaced data
– many different measures (traditional semivariogram, non-ergodic
covariance / correlogram, relative (general and pairwise) variograms,
madogram/rodogram, indicator)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Programs in GSLIB (2)
• Kriging:
– 1,2 or 3-D grid kriging, cross validation, jackknifing
– SK, OK, UK, kriging with external drift
– cokriging
– indicator kriging
• Stochastic Simulation:
– Gaussian methods (LU matrix method, sequential)
– Indicator methods (cosimulation, Markov Bayes)
– Boolean simulation
– Simulated annealing
– P-field simulation

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
GSLIB Preliminaries (1)
• The latest Stanford geostatistical algorithms
• All source code is (almost) ANSI Standard FORTRAN 77
• Tested on IBM compatible PC's, SGI, DEC, SUN, HP ...
• Requirements: FORTRAN compiler, basic programming skills, and some idea
of what you are after.
• Complete with main programs which all require a prepared parameter file.
• Intended to be the foundation for custom programs.

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
GSLIB Preliminaries (2)
• Input/Output files are ASCII flat files compatible with Geo-EAS and the
Geostatistical Toolbox:

2d data file example:


data2dg.dat) File name
4 Number of columns

}
Xlocation
Ylocation
Column names
ATTRIBUTE
COVARIATE
28.5 23.5 3.852 1.003
30.5 45.5 5.543 1.772
22.5 18.5 .102 .327
49.5 35.5 4.548 2.833
} Space/Tab delimited data

• Graphical output is PostScript page description language (GhostScript would


be a useful supplement).

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Grid Specification
• The GSLIB rectangular grid is denoted as:

• Could be associated to any Cartesian or stratigraphic coordinate system.


• Output grids are ordered with x cycling fastest, then y, and then z. Index
location of node ix,iy,iz is located by:
loc = (iz-1)*nx*ny + (iy-1)*nx + ix
• 2-D grids are specified with nz=1. zmn and zsiz are not used.
• Grids or simulated realizations are ordered one after another.
• Some sort of compressed output format may be necessary.
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Example Parameter File
Parameters for GAMV
*******************
START OF PARAMETERS:
../ data/ clus ter.dat \ file w ith data
1 2 0 \ columns for X, Y, Z coord ina tes
2 3 4 \ numb er of varab les ,column numb ers
-1.0e21 1.0e21 \ trimming limits
g amv.out \ file for variog ra m outp ut
10 \ numb er of lag s
5.0 \ lag s ep ara tion d is tance
3.0 \ lag tolerance
3 \ numb er of d irections
0.0 90.0 50.0 0.0 90.0 50.0 \ azm,atol,b and h,d ip ,d tol,b and v

...
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Reservoir Modeling with GSLIB

Data Preparation and Statistical


Displays
• Data Cleaning / Quality Control
• Statistics as Parameters for Random Function Models
• Univariate Statistics
• Histograms and Probability Plots
• Q-Q and P-P Plots
• Bivariate Statistics and Distributions
• Exploratory Data Analysis

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Statistical Analysis of Data
Well Depth X Y Poro Perm Layer Seq LogK
10100 7158.9 672.7 2886.0 27.00 554.000 1 51 2.744
10100 7160.4 672.8 2886.0 28.60 1560.000 1 51 3.193
10100 7160.7 672.8 2886.0 30.70 991.000 1 51 2.996
10100 7161.0 672.8 2886.0 28.40 1560.000 1 51 3.193
10100 7161.9 672.9 2886.0 31.00 3900.000 1 51 3.591
10100 7162.3 672.9 2886.0 32.40 4100.000 1 51 3.613

• Goals of Exploratory Data Analysis
– understand the data: statistical versus geological populations
– ensure data quality
– condense information
• Only limited functionality in GSLIB (some special geostat tools) a supplement with
commercial software

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Statistics as Parameters of
Random Function Models

• Not interested in sample statistics ➾ need to access underlying population parameters


• A model is required to go beyond the known data
• Because earth science phenomena involve complex processes they appear as random.
Important to keep in mind that actual data are not the result of a random process.

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Univariate Description

Frequency

• Histogram: A counting of samples in classes


• Sometimes two scales are needed to show the details (use trimming limits)
• logarithmic scale can be useful
• Summary statistics
– mean is sensitive to outliners
– median is sensitive to gaps in the middle of a distribution
– locate distribution by selected quantiles (e.g., quartiles)
– spread measured by standard deviation (very sensitive to extreme values)
• GSLIB program his tp lt

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Cumulative Probability Cumulative Probability Plot

0.01 0.10 1.00 10.0

Variable

• Useful to see all of the data values on one plot


• Useful for isolating statistical populations
• May be used to check distribution models:
– straight line on arithmetic scale a normal distribution
– straight line on logarithmic scale a lognormal distribution
– small departures can be important
– possible to transform data to perfectly reproduce any univariate distribution
• GSLIB program probplt
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Cumulative Histograms

Cumulative Frequency
1
Frequency

The cumulative frequency is the total or the cumulative fraction of samples less
than a given threshold

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Cumulative Histograms
1.0

Cumulative Frequency 0.25

0.0
First Quartile
Value
• Cumulative frequency charts do not depend on the binwidth; they can be
created at the resolution of the data
• A valuable descriptive tool and used for inference
• A quantile is the variable-value that corresponds to a fixed cumulative
frequency
– first quartile = 0.25 quantile
– second quartile = median = 0.5 quantile
– third quartile = 0.75 quantile
can read any quantile from the cumulative frequency plot
• Can also read probability intervals from the cumulative frequency plot (say,
the 90% probability interval)
• Direct link to the frequency
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Q-Q / P-P Plots
Q-Q Plot: Equal Weighted P-P Plot: Equal Weighted
True Value 20.0 1.0

True Value
10.0 0.5

0.0 10.0 20.0 0.0 0.5 1.0


Clustered Data Clustered Data

• Compares two univariate distributions


• Q-Q plot is a plot of matching quantiles a a straight line implies that the two
distributions have the same shape.
• P-P plot is a plot of matching cumulative probabilities a a straight line
implies that the two distributions have the same shape.
• Q-Q plot has units of the data, P-P plots are always scaled between 0 and 1
• GSLIB program q p plt

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Q-Q Plots
Da ta Set One Da ta Set Tw o
value cd f value cd f
0.010 0.0002 0.060 0.0036
0.020 0.0014 0.090 0.0250
0.020 0.0018 0.090 0.0321
0.020 0.0022
0.030 0.0034
0.030 0.0038
0.960 0.4998 2.170 0.4964
0.960 0.5002 2.220 0.5036
38.610 0.9962
40.570 0.9966
42.960 0.9978
43.500 0.9982 19.440 0.9679
46.530 0.9986 20.350 0.9750
102.700 0.9998 58.320 0.9964

• Sort the values in each data set


• Calculate cumulative distribution function (CDF) for each
• Match according to (CDF) values
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Let's Build a Q-Q Plot

Frequency
Frequency

Cumulative Frequency
Cumulative Frequency

Core Porosity Log Porosity

• Histograms of core porosity and log-derived porosity


• Preferential sampling explains difference; these are not “paired” samples so we can not
detect bias in samples

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Let's Build a Q-Q Plot

Log porosity

Core porosity

• Read corresponding quantiles from the cumulative frequency plots on the previous page
• Plot those quantiles on the plot

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Univariate Transformation
Cumulative Frequency

Cumulative Frequency
Frequency
Frequency

• Transforming values so that they honor a different histogram may be done by matching
quantiles
• Many geostatistical techniques require the data to be transformed to a Gaussian or
normal distribution

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Data Transformation
Core Porosity Core Porosity

Cumulative Frequency
Frequency

Log-derived Porosity Log-derived Porosity

Cumulative Frequency
Frequency

• Histograms of core porosity and log-derived porosity


• These are “paired” samples so we may want to transform the log-derived porosity
values to core porosity

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Data Transformation
• Use the cumulative frequency plots on the previous page to fill in the following table

Log Porosity Core Porosity


10.0 7
15.0 10
20.0 18
25.0 26
30.0 29
• Could we transform 20000 log-derived porosity values according to the 853
paired samples we have?
• Under what circumstances would we consider doing this?
• What problems might be encountered?

Core Porosity Log-derived Porosity


Cumulative Frequency
Cumulative Frequency

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Monte Carlo Simulation

Cumulative Frequency
Frequency
0.7807

28.83
Core Porosity Core Porosity

• Monte Carlo Simulation / Stochastic Simulation / Random Drawing proceed


by reading quantiles from a cumulative distribution
The procedure:
• generate a random number between 0 and 1 (calculator, table, program, ...
• read the quantile associated to that random number
For Example:
Random Number Simulated Number
0.7807 28.83 ...
0.1562
0.6587
0.8934

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Scatterplots
True versus Estimate

16.0

True Value ρrank > ρ


0.0 16.0
Estimate

• Bivariate display, estimate-true, two covariates, or the same


variable separated by some distance vector
• Linear correlation coefficient ranges between -1 and +1 and
ρrank < ρ
is sensitive to extreme values (points away from the main cloud)
• Rank correlation coefficient is a useful supplement:
– if ρrank > ρ then a few outliers are spoiling an otherwise good correlation
– if ρrank < ρ then a few outliers are enhancing an otherwise poor correlation
– if ρrank = 1 then a non-linear transform of one covariate can make ρ = 1
• GSLIB program s catplt

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Scatterplots

• Look at bivariate summaries


• Marginal histograms

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Bivariate Distributions

Bivariate histogram
a)

Bivariate cumulative
b)
distribution function

Conditional cumulative
c)
distribution function

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Conditional Distributions
Permeability (md) Distribution of possible
permeability values at a
known porosity value

Known primary value

Calibration Scattergram
z - values

• Prediction of conditional distributions is at the heart of geostatistical algorithms

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Class Definition
10,000

Permeability

0.01
0.0 0.20

Porosity
• Choose equal probability intervals rather than equal porosity or permeability intervals
• Check for bias in histogram - often there are many more low porosity log data than low
porosity core data
• Due to a biased core data set, we may need to set porosity cutoffs based on an even φ
spacing rather than equal probability

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Exploratory Data Analysis
• Plot the data in different ways; our eyes are good at pattern detection
• Choose geological/statistical populations for detailed analyses:
– populations must be identifiable in wells without core
– must be able to map these populations (categories)
– can not deal with too many, otherwise there are too few data for reliable
statistics
– often a decision must be made to pool certain types of data
– stationarity is a property of statistical models and not reality
– important and very field/data/goals-specific
• Perform statistical analyses within each population:
– ensure data quality
– look for trends
– understand “physics” as much as possible
• Decluster data for geostatistical modeling
• Statistical tools are used throughout a reservoir characterization study

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Reservoir Modeling with GSLIB

Variogram Calculation and


Interpretation
• Spatial Statistics
• Coordinate and Data Transformation
• Define the Variogram
• How to Calculate Variograms
• “Visual Calibration”
• Variogram Interpretation
• Show Expected Behavior
• Work Through Some Examples
• Test Your Understanding

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Spatial Statistics

• Spatial variability/continuity depends on the detailed distribution of the


petrophysical attribute; our measure must be customized for each field and
each attribute (φ,Κ)
• Depending on the level of diagenesis, the spatial variability may be similar
within similar depositional environments. The recognition of this has led to
outcrop studies.

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Data Transformation
Why do we need to worry about data transformation?
• Attributes, such as permeability, with highly skewed data distributions present
problems in variogram calculation; the extreme values have a significant
impact on the variogram.
• One common transform is to take logarithms,
y = log10 ( z )
perform all statistical analyses on the transformed data, and back transform at
the end → back transform is sensitive
• Many geostatistical techniques require the data to be transformed to a
Gaussian or normal distribution.
The Gaussian RF model is unique in statistics for its extreme analytical
simplicity and for being the limit distribution of many analytical theorems
globally known as “central limit theorems”

The transform to any distribution (and back) is easily accomplished by the quantile transform

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Normal Scores Transformation
• Many geostatistical techniques require the data to be transformed to a
Gaussian or normal distribution:
Cumulative Frequency

Cumulative Frequency
Frequency
Frequency

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Definition of the Variogram

Variogram, γ(h)
No correlation
u + h)

La
g
Ve
Vector (

c
to
r(

Increasing
h)

Variability
Location

atio n Vector (u)


Loc

Origin Lag Distance (h)


• In probabilistic notation, the variogram is defined as:
• 2 γ (h ) = E{[ Z(u ) − Z(u + h )]2 } - for all possible locations u
• The variogram for lag distance h is defined as the average squared difference
of values separated approximately by h:
1
2 γ (h ) = ∑
N(h ) N ( h )
[z(u ) − z(u + h )]2
where N(h) is the number of pairs for lag h

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Variogram Calculation
• Consider data values separated by lag vectors

ρ 0.81 ρ 0.77
γ0.19 γ0.23

Tail
Tail

Head Head

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Spatial Description
The Variogram is a tool that Quantifies Spatial Correlation

γ γ γ

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Calculating Experimental
Variograms
• 2-D or 3-D, regular or irregular spaced
• Direction specification (regular):

• Direction specification (irregular):


Y axis (North)

ce
an
ler
g To
La Bandwidth
4
Azimuth Lag
3
Lag
ce
stan
i 2
g D Lag Azimuth
La
tolerance
1
Lag
X axis (East)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Calculating Experimental
Variograms
Example: Starting With One Lag (i.e. #4)

1
2γ ( h) = ∑ [ z ( u) − z ( u + h)]2
N ( h) N ( h )

Start at a node, and compare


value to all nodes which fall in
the lag and angle tolerance.
...

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Calculating Experimental
Variograms
1
2γ ( h) = ∑ [ z ( u) − z ( u + h)]2
N ( h) N ( h )

Move to next node.

...

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Calculating Experimental
Variograms
Now Repeat for All Nodes
And Repeat for All Lags
Variogram, γ(h)

No correlation

Increasing
Variability
...

Lag Distance (h)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Variogram Calculation Options
• Data variable (transformed?) and coordinates (transformed?)
• Number of directions and directions:
– compute the vertical variograms in one run and the horizontal variograms
in another
– often choose three horizontal directions: omnidirectional, “major”
direction, and perpendicular to major direction
– azimuth angles are entered in degrees clockwise from north
• Number of lags and the lag separation distance:
– lag separation distance should coincide with data spacing
– the variogram is only valid for a distance one half of the field size a
choose the number of lags accordingly
• Number and type of variograms to compute:
– there is a great deal of flexibility available, however, the traditional
variogram applied to transformed data is adequate in 95% of the cases
– typically consider one variogram at a time (each variogram is computed
for all lags and all directions)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Interpreting Experimental
Variograms
Vertical Variogram
Sill

Nugget Effect
Range

Distance
• sill = the variance (1.0 if the data are normal scores)
• range = the distance at which the variogram reaches the sill
• nugget effect = sum of geological microstructure and measurement error
– Any error in the measurement value or the location assigned to the
measurement translates to a higher nugget effect
– Sparse data may also lead to a higher than expected nugget effect

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Challenges in Variogram
Calculation
• Short scale structure is most important
– nugget due to measurement error should not be modeled
– size of geological modeling cells
• Vertical direction is typically well informed
– can have artifacts due to spacing of core data
– handle vertical trends and areal variations
• Horizontal direction is not well informed
– take from analog field or outcrop
– typical horizontal vertical anisotropy ratios

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Interpreting Experimental
Variograms
Vertical Variogram
Sill

Distance

• vertical permeability variogram


• sill: clearly identified (variance of log Κ data)
• nugget: likely too high

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Trend
Example Trend Data Set Vertical Variogram

3.0

γ Sill
Vertical

0.0

-3.0

Distance
Horizontal

• indicates a trend (fining upward, …)


• could be interpreted as a fractal
• model to the theoretical sill; the data will ensure that the trend appears in the
final model
• may have to explicitly account for the trend in later simulation/modeling

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Cyclicity
Example Cyclic Data Set Vertical Variogram
Sill
3.0

γ
Vertical

0.0

-3.0

Horizontal Distance

• cyclicity may be linked to underlying geological periodicity


• could be due to limited data
• focus on the nugget effect and a reasonable estimate of the range

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Geometric Anisotropy
Example Geometric Anisotropy Data Set
Vertical Variogram Sill
3.0

γ
Vertical

0.0 Horizontal Variogram

-3.0
Distance (h)
Horizontal

• Compare vertical sill with horizontal sill


• When the vertical variogram reaches a higher sill:
– likely due to additional variance from stratification/layering
• When the vertical variogram reaches a lower sill:
– likely due to a significant difference in the average value in each well a
horizontal variogram has additional between-well variance
• There are other explanations
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Zonal Anisotropy
Example Zonal Anisotropy Data Set
Sill
3.0 Vertical Variogram

γ Apparent Sill
Vertical

Horizontal Variogram
0.0

-3.0
Distance (h)
Horizontal

• Compare vertical sill with horizontal sill


• When the vertical variogram reaches a higher sill:
– likely due to additional variance from stratification/layering
• When the vertical variogram reaches a lower sill:
– likely due to a significant difference in the average value in each well a
horizontal variogram has additional between-well variance
• There are other explanations

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Horizontal Variograms
A few experimental horizontal variograms:
Horizontal: Layer 01 Horizontal: Layer 13
γ γ

Distance Distance

Horizontal: Layer 14 Horizontal: Sand Horizontal: Shale

Distance Distance Distance


Noise is often due to scarcity of data in the horizontal direction.

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Variogram Interpretation and
Modeling
Key is to apply geologic knowledge to the experimental variogram and to build a legitimate
(positive definite) variogram model for kriging and simulation (discussed later)

This ensures:
• that the covariance can be assessed over all lag vectors, h.
• that the variogram will be a legitimate measure of distance
The sum of known positive definite models is positive definite. There is great
flexibility in modeling variograms with linear combinations of established
models.
Some common positive definite models:

Nugget Effect Spherical Exponential Gaussian

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Horizontal Variograms

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Porosity Variogram
Vertical Variogram Horizontal Variogram

γ γ

type sill Range type sill Range


spherical 0.4 1.5 spherical 0.4 500.0
spherical 0.6 15..3 spherical 0.6 4000.0

Distance Distance

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Summary
• Variogram is very important in a geostatistics study
• Measure of geological distance with respect to Euclidian distance
• Initial coordinate and data transformation
• Calculation principles
• Interpretation principles:
– trend
– cyclicity
– geometric anisotropy
– zonal anisotropy
• Variogram modeling is important (experimental points are not used)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Reservoir Modeling with GSLIB

Kriging
• Weighted Linear Estimators
• Some Definitions
• Derivation of the Kriging Equations
• Some Examples
• Different Types of Kriging
• How / Where is Kriging Used

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Weighted Linear Estimators
• The basic idea is to estimate the attribute value (say, porosity) at a location
where we do not know the true value
n
Z ( u ) = ∑ λ i ⋅ Z( u i )
*

i =1
• where u refers to a location, Z*(u) is an estimate at location u, there are n data
values Z(ui), i=1,...,n, and λi refer to weights.

• What factors could be considered in assigning the weights?


–closeness to the location being estimated
–redundancy between the data values
–anisotropic continuity (preferential direction)
–magnitude of continuity / variability
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Weighted Linear Estimators

• Assign all of the weight to the nearest data (polygonal-type estimate)


• Assign the weights inversely proportional to the distance from the location
being estimated (inverse distance schemes)
1
c + d iw
λi =
1
∑ i =1 c + d w
n

where di is the distance between data i andi the location being


estimated, c is a small constant, and ω is a power (usually between 1 to 3).
• How about using the variogram? a that is kriging

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Some Definitions
• Consider the residual data values:
Y(ui)= Z(ui) - m(ui), i=1,…,n
• where m(u) could be constant, locally varying, or considered constant but
unknown.
• Variogram is defined as:
2 γ(h) = E{[ Y(u}) - Y(u + h]2}
• Covariance is defined as:
C(h) = E{ Y(u) •Y(u + h)}
• Link between the Variogram and Covariance:
2 γ(h) = [ E{ Y2(u) + [ E{ Y2(u + h)}] - 2 • [ E{ Y(u) • Y(u + h)]
= Var{Y(u)} + Var{Y(u + h)} - 2 • C(h})
= 2 [ C(0) - C(h)]
So, C(h) = C(0) - γ(h)
C(0) = sill
γ(h)
C(h)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Simple Kriging (1)
• Consider a linear estimator:
n
Y (u ) = ∑ λ i ⋅ Y(u i )
*

i =1
• where Y(ui) are the residual data (data values minus the mean) and Y*(u) is
the estimate (add the mean back in)
• The error variance is defined as
E{[Y* (u ) − Y(u )]2 } A2-2ab+b2
E {[Y * ( u)]2 } − 2 ⋅ E {Y * ( u) ⋅ Y ( u)} + E {[Y ( u)]2 }
n n n

∑ ∑ λ λ E{Y ( u ) ⋅ Y ( u )} − 2 ⋅ ∑ λ E {Y ( u) ⋅ Y ( u )}
i = 1 j =1
i j i j
i =1
i i + C ( 0)

n n n

∑ ∑ λ λ C ( u , u ) − 2 ⋅ ∑ λ C ( u, u )
i = 1 j =1
i j i j
i =1
i i
+ C (0)
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Simple Kriging (2)
• Optimal weights λi,i=1,…,n may be determined by taking partial derivatives of
the error variance w.r.t. the weights
∂[ ] n
= 2 ⋅ ∑ λ jC(u i , u j ) − 2 ⋅ C(u, u i ) , i = 1,..., n
∂λ i j= 1
• and setting them to zero
n

∑ λ C(u , u ) = C(u, u ) ,
j= 1
j i j i i = 1,..., n

• This system of n equations with n unknown weights is the simple kriging


(SK) system

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Simple Kriging: Some Details
γ 2,3
γ 1,2 γ 1,3

γ 0,1 γ 0,2
γ 0,3

• There are three equations to determine the three weights:


λ 1 ⋅ C(1,1) + λ 2 ⋅ C(1,2) + λ 3 ⋅ C(1,3) = C(0,1)
λ 1 ⋅ C( 2,1) + λ 2 ⋅ C( 2,2) + λ 3 ⋅ C( 2,3) = C(0,2)
λ 1 ⋅ C( 3,1) + λ 2 ⋅ C( 3,2) + λ 3 ⋅ C( 3,3) = C(0,3)

• In matrix notation: (Recall that C(h}) = C(0) - γ( h))


 C(1,1) C(1,2) C(1,3)   λ 1   C(0,1) 
C( 2,1) C( 2,2) C( 2,3) λ  = C(0,2)
  2   
C( 3,1) C( 3,2) C( 3,3) λ 3  C(0,3)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Simple Kriging
Changing the Range

range = 1 range = 5

γ range = 10

Distance

Simple kriging with a zero nugget effect and an isotropic spherical variogram with
three different ranges:

λ1 λ 2 λ 3
range = 10 0.781 0.012 0.065
5 0.648 -0.027 0.001
1 0.000 0.000 0.000

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Simple Kriging
Changing the Nugget Effect
100%
75%
γ
nugget = 25%

Distance

Simple kriging with an isotropic spherical variogram with a range of 10


distance units and three different nugget effects:

λ1 λ 2 λ 3
nugget = 0% 0.781 0.012 0.065
25% 0.468 0.203 0.064
75% 0.172 0.130 0.053
100% 0.000 0.000 0.000

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Simple Kriging
Changing the Anisotropy

Simple kriging with a spherical variogram with a nugget of 25%, a principal range
of 10 distance units and different “minor” ranges:

λ1 λ 2 λ 3
anisotropy 1:1 0.468 0.203 0.064
2:1 0.395 0.087 0.141
5:1 0.152 -0.055 0.232
20:1 0.000 0.000 0.239

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada

Kriging
All versions of kriging are elaborations on the basic linear regression
algorithm and corresponding estimator:
n
[Z (u ) − m(u )] = ∑ λ α (u )[Z(u α ) − m(u α )]
*
SK
• α =1
where Z(u) is the RV model at location u, the uα 's are the n data locations,
m(u) = E{Z(u) is the location-dependent expected value of RV Z(u), and
ZSK*(u) is the linear regression estimator, also called the “simple kriging”
(SK) estimator.
• The SK weights λα (u) are given by the general non-stationary normal
equations:
n



Some Remarks:
λ
β =1
β ( u)C ( uβ , uα ) = C ( u, uα ) , α = 1,...., n
– there are many types of kriging where specific constraints or methods of
application are considered
– the weights λα (u) account for (1) the proximity of the data to the location
being estimated and (2) the clustering of the data
– traditionally used for mapping
– modern use is in the construction of local distributions of uncertainty for
stochastic simulation algorithms

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Kriging
• Minimum error variance estimate of the unknown:

z (u ) = ∑ λ α (u )Z(u α ) + 1 − ∑ λ α (u )m
n n
*

α =1
 α =1 
• Simple kriging (SK): no constraint, the mean m is known.
• Ordinary kriging (OK): the sum of the weights ∑nα=1 λα (u) constrained to
equal 1.
• Kriging with a trend model (KT) considers that m is unknown and that it has
a more complex trend of known shape but unknown parameters.
L
m(u ) = ∑ a l f l (u )
l =0

• where m(u) is the local mean, a1, l = 0...,L are unknown coefficients of the
trend model, and fl(u) are low order monomials of the coordinates. GSLIB
allows nine different monomials up to quadratic functions of the coordinates
( x , y , z , xx , yy , zz, xy, xz , and, yz ).
• Kriging with an external drift is an extension of KT. Considers a single trend
function f1(u) defined at each location from some external (secondary)
variable.
• Cokriging is for kriging with different types of data.
• Indicator Kriging has a different goal a to build a conditional cdf rather than
derive an estimate.

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
G
eo
s t
a
va tist
A
G
ila ic SL
bl al
ea So IB
tw ftw
Parameter File For KT2D Parameters for KB2D
w
w
.G re
LS
a
LI
IB Br
******************* .co ar
m y
START OF PARAMETERS:
../data/cluster.dat \ file with data
1 2 3 \ columns for X, Y, and variable
-1.0e21 1.0e21 \ trimming limits
3 \ debugging level: 0,1,2,3
kb2d.dbg \ file for debugging output
kb2d.out \ file for kriged output
5 5.0 10.0 \ nx,xmn,xsiz
5 5.0 10.0 \ ny,ymn,ysiz
1 1 \ x and y block discretization
4 8 \ min and max data for kriging
20.0 \ maximum search radius
1 2.302 \ 0=SK, 1=OK, (mean if SK)
1 2.0 \ nst, nugget effect
1 8.0 0.0 10.0 10.0 \ it, c, azm, a_max, a_min

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
G
eo
s t
a
va tist
A
G
ila ic SL
bl al
ea So IB
tw ftw
Parameter File For KT3D Parameters for KT3D
w
w
.G re
LS
a

IB
LI
Br
******************* .co ar
m y
START OF PARAMETERS:
../data/cluster.dat \ file with data
1 2 0 3 4 \ columns for X, Y, Z, var, sec var
-1.0e21 1.0e21 \ trimming limits
1 \ option: 0=grid, 1=cross, 2=jackknife
xvk.dat \ file with jackknife data
1 2 0 3 0 \ columns for X,Y,Z,vr and sec var
3 \ debugging level: 0,1,2,3
kt3d.dbg \ file for debugging output
kt3d.out \ file for kriged output
50 0.5 1.0 \ nx,xmn,xsiz
50 0.5 1.0 \ ny,ymn,ysiz

...
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
G
eo
s t
a
va tist
A
G
ila ic SL
bl al
ea So IB
tw ftw
Parameter File For KT3D w
w
.G re
LS
a

IB
LI
Br
.co ar
1 0.5 1.0 \ nz,zmn,zsiz m y
1 1 1 \x,y and z block discretization
4 8 \ min, max data for kriging
0 \ max per octant (0-> not used)
20.0 20.0 20.0 \ maximum search radii
0.0 0.0 0.0 \ angles for search ellipsoid
0 2.302 \ 0=SK,1=OK,2=non-st SK,3=exdrift
000000000 \ drift: x,y,z,xx,yy,zz,xy,xz,zy
0 \ 0, variable; 1, estimate trend
extdrift.dat \ gridded file with drift/mean
4 \ column number in gridded file
1 0.2 \ nst, nugget effect
1 0.8 0.0 0.0 0.0 \ it,cc,ang1,ang2,ang3
10.0 10.0 10.0 \ a_hmax, a_hmin, a_vert

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Kriging
• Kriging is a procedure for constructing a minimum error variance linear
estimate at a location where the true value is unknown
• The main controls on the kriging weights are:
– closeness of the data to the location being estimated
– redundancy between the data
– the variogram
• Simple Kriging (SK) does not constrain the weights and works with the
residual from the mean
• Ordinary Kriging (OK) constrains the sum of the weights to be 1.0, therefore,
the mean does not need to be known
• There are many different types of kriging
• Two implicit assumptions are stationarity (work around with different types of
kriging) and ergodicity (more slippery)
• Kriging is not used directly for mapping the spatial distribution of an attribute
(sometimes when the attribute is smooth). It is used, however, for building
conditional distributions for stochastic simulation

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Reservoir Modeling with GSLIB

Some Practical Aspects of


Reservoir Modeling
• Laws of Heterogeneity / Why Build Geologic Models?
• Steps in a Geostatistical Reservoir Modeling Study
• Hierarchical Approach to Modeling
• Spatial Continuity
• Estimation / Mapping of Petrophysical Properties
• Stochastic Simulation / Modeling
• Issues and Concerns

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Preliminaries (1)
• Laws or Heterogeneity:
1. All reservoirs are heterogeneous
2. All reservoirs are more heterogeneous than first imagined
3. The degree of heterogeneity is directly proportional to the amount of time
allocated for the project and the project funding
• Why Build 3-D Geologic Models?
– handle large amounts of data
– consistent analysis in three dimensions
– direct numerical input to flow simulation and pore volume calculation
(reservoir management)
– test / visualize multiple geologic interpretations
– assess uncertainty

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Preliminaries (2)
• Historical Perspective of Geostatistical Modeling:
– Theory of probability (in its modern form) was formalized in the 1600's
by Blaise Pascal and Pierre de Fermat. Others: Bayes, Gauss, ...
– The foundation for geostatistical techniques was established by people
like Kolmogorov, Weiner, Matern, and Gandin in the early 1900's
– Geostatistics was started in the 1960's by Krige and Sichel in South Africa
and Matheron in France. Two of Matheron's first students (Journel and
David) would start new centers of teaching and research in the USA and
Canada
– Application became popular in mining and meteorology. Now, these
techniques are applied in many fields from fisheries, forestry,
environmental remediation, and so on
– Extensively used by major oil companies

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Some of the Data Available
for Reservoir Modeling
Data Integration is a fundamental principle of geostatistics / reservoir modeling; the
goal is to explicitly account for all of the available data. A large part of the ongoing
research in Geostatistical Reservoir Modeling is to devise techniques that can
accommodate a greater variety of data. Following are some of the data that are
considered:
• Well Log Data (surface tops, rock type, φ, Κ) by zone
• Core Data (φ and Κ by rock type) by zone
• Sequence Stratigraphic Interpretation / Layering (a definition of the continuity and
trends within each layer of the reservoir)
• Trends and Stacking Patterns available from a regional geological interpretation
• Analog data from outcrops or densely drilled similar fields (size distributions,
measures of lateral continuity)
• Seismic-Derived Attributes (vertically averaged rock type proportions and porosity)
• Well Test and Production Data (interpreted K • thickness, interpreted channel widths,
connected flow paths, barriers)
This information is sparse relative to the size of the heterogeneities being modeled;
therefore, there is always uncertainty in the geological model

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Constructing 3-D Models
The specific process employed for 3-D model building will depend on the data available, the
time available, the type of reservoir, and the skills of the people available. In general, the
following major steps are required:
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 the spatial correlation of the rock
types
(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
(g) merge and translate back to real coordinate space
4. Verify the model
5. Combine zones into a single model
• Each of these steps is addressed during this lecture or course (to some extent)
• Uncertainty is assessed by deriving reasonable estimates of uncertainty for each input
parameter and then generating multiple realizations

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Conceptual Geological
Model / Zone Definition

Select zones by considering:


• sequence stratigraphic zonation
• keep geologically “homogeneous” rock together
• maintain a reasonable number of data per zone
• less resolution in water bearing formation

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Geological Correlation Style
Each layer in the reservoir is classified as belonging to one of the
following geological correlation styles. The existing grids defining
the zone and the restored grids are used for modeling.

• Proportional (conforms to existing top and base):

• Truncation (conforms to existing base):

• Onlap (conforms to existing top):

• Offlap (does not conform to existing top or base):

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Stochastic Modeling of Surfaces
To assess uncertainty in pore volume or reservoir performance predictions requires
adding uncertainty to the gridded surface elevations. Some characteristics of the
uncertainty:
• essentially zero at the well locations
• varies smoothly away from the wells
• variance depends on the quality of the seismic and the distance from the wells

The following example was created by using sgsim to create a correlated Gaussian
error. A section through four wells:

Multiple surfaces would be handled by constructing stochastic isochore maps and


subtracting them from the top surface. The grids are constrained to honor the well
data, not cross each other, and fall within realistic bounds of uncertainty.

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Areal and Vertical Trends
• Often it is possible to infer areal or vertical trends in the distribution of rock
types and/or petrophysical properties.

20.0

Stratigraphic Position
Northing

10.0

Easting
Mean Porosity

• These trend profiles/maps may be used in many geostatistical modeling programs


• Trends should be removed and residuals are modeled stochastically

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Declustering
Data may be clustered in high pay zones or in certain areas; a declustering
procedure is required to assign relative weights. These declustering weights can
then be used when looking at histograms or summary statistics or when model
building.
Given an example with eight values. With cell declustering, the relative
declustering weight for each datum would be:

Well Weight
W-1 1.2
W-2 1.13
W-3 0.8
W-4 0.85
W-5 0.74
W-6 0.93
W-7 1.1
W-8 1.26

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Histogram Smoothing / Modeling
Sparse data may create a need to smooth or model the histogram of the attribute under
consideration. Two examples, one with eight data values and an example with 243 core
permeability measurements from a bimodal distribution:

Field Average
Frequency

Frequency

Net-to-Gross from Well

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Definition of the Variogram
A quantitative measure of spatial variability/continuity is needed to characterize
the detailed distribution of attributes within the reservoir; this measure must be
customized for each field and each attribute (φ,Κ)
The variogram is one way to quantify spatial variability:

γ(h)
Lag Distance (h)
• The variogram for lag distance h is defined as the average squared difference
of values separated approximately by h:
1
2 γ (h ) = ∑
N(h ) N ( h )
[ z ( u ) − z ( u + h )]2

where N(h}) is the number of pairs for lag h.


• In probabilistic notation, the variogram is defined as:

2 γ (h ) = E{[Z(u ) − Z(u + h )]2 }

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Spatial Information for Object-
based Modeling (1)
Object-based modeling techniques require information on the size, shape, and relationship
between the different objects. For braided fluvial reservoirs, some of the needed
information includes:

• fraction of channel sand (could vary areally and vertically)


• width and thickness of channel sands (could vary vertically and follow a
distribution of possible sizes)
• measures of channel sinuosity (depend on size of channel and vertical
position)
• geometry of channel “families” or multi-story channels

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Spatial Information for Object-
based Modeling (2)
An example of how the channel size (say thickness) could be specified:

The vertical axis on this plot represents restored stratigraphic position and the
horizontal axis is channel thickness. The Q1, Q2, and Q3 lines represent the
quartiles (25%, 50% and 75%) values of the distribution. Note how the channels
are smaller at the top of the zone.

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Sequential Indicator Simulation
• Generate indicator-based realizations that reproduce local conditioning data,
global proportions, local proportions (via locally varying proportions), and
patterns of spatial correlation (variogram)
• Define an indicator transform:

i ( uα ; k ) = 1, i f
0, i f
lithosfacie s k
not
presen t at locatio n uα

• Generate a 3-D realization of that indicator variable


• Example from the Ghawar field in Saudi Arabia:

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Object-Based Modeling
• Simulate the deposition of the reservoir by stochastically positioning
geometric shapes
• Start from the bottom and alternately lay down floodplain sediments and
channel fill. Specify the distribution of channel sizes, spacing, and so on.
• Could additionally model crevasse deposits and point bar sands
• Honor limited well data by controlling the channel positions
• Commonly applied to fluvial reservoirs

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Sequential Gaussian Simulation
• A technique that is robust and applicable for the generation of realizations of
continuous variables. The realizations can be made to honor:
– local conditioning data,
– the global histogram (declustered and smoothed),
– areal and vertical trends (via locally varying mean), and
– patterns of spatial correlation (variogram)
• Works with a Gaussian or Normal transform of the data (see Normal Scores
Transformation)
• Generate a 3-D realization of Gaussian variable and back transform
• Applied on a by zone and by rock type basis between the restored grids for
geological correlation

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Annealing Cosimulation
• A technique that is robust and applicable for the generation of realizations of
continuous variables (specifically permeability). The realizations can be made
to honor:
– local conditioning data,
– the global histogram (declustered and smoothed),
– areal and vertical trends (via locally varying mean), and
– patterns of spatial correlation (variograms and indicator variograms -- for
special continuity of high and low values)
– a cross plot of porosity and permeability
• The stochastic simulation problem is posed as an optimization problem and the
simulated annealing algorithm is used to solve the problem, i.e., generate
plausible realizations
• Applied on a by zone and by rock type basis between the restored grids for
geological correlation

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Criteria for Model
Verification
• Many interdependent subjective decisions are made in the construction of a
geostatistical reservoir model. Some things that should be checked:
– does the model appear geologically plausible?
– relative to other models, are the heterogeneities reasonably distributed?
– are the porosity and permeability models consistent with the rock type model?
– is the geological correlation style correct (mistake with Zrel)?
– does the model present trends consistent with the regional geology?
• Does the model honor all of the input data?
– local conditioning data (plot the results),
– declustered histogram (Q-Q plot),
– variograms,
– cross plot between porosity and permeability,
• Do the techniques employed pass all of the cross-validation checks?
• Can the model be checked relative to data that were not used in the model building,
e.g., well test or production history?

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Q-Q Plots to Assess
Histogram Reproduction
• A Q-Q plot is to compare two histograms or univariate distributions, e.g., the
input declustered histogram of porosity for a specific rock type within a
specific zone and the distribution from the final 3-D model
• A plot of the matching quantiles of two distributions. For example, one point
on the plot is the median of the first distribution plotted against the median of
the second distribution. If the points for many quantiles (1%, 2%, ..., 98%,
99%) fall on a straight line then the two distributions agree
• The two axes are in units of the data. The following is an actual example from
one zone of a large clastic reservoir:
1/100 of simulated values

Data distribution
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Assessing Variogram
Reproduction
• Plot of the input variogram model (solid line) with the variogram from the 3-D
realization (black dots) in three major directions:

γ γ

Distance (m) Distance (m)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Assessing Variogram
Reproduction
• A 2-D example
• The variogram should be reproduced on average
Model, Average and Experimental Variograms
(10 Realizations)

γ (h)
1.0

Average variogram
Model variogram
Experimental Variograms (North)
Experimental Variograms (East)

0.0 25.0
Distance (m)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Summary
• All reservoirs are heterogeneous
• Reasons for building 3-D models (data integration, refine estimates of PV,
quantify uncertainty, assess continuity, quantify uncertainty in predictions,...)
• Procedure for modeling:
– geological zonation, layering, conceptual model
– statistics: declustering, modeling, variograms
– rock type modeling (indicator, object-based, hybrid)
– porosity modeling
– permeability modeling
– model validation
• Issues that have been glossed over:
– size scaling from core to geological modeling cell
– faults and fractures
– reliable inference of spatial statistics
– hierarchical modeling

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Reservoir Modeling with GSLIB

Indicator Simulation for


Categorical Variables
• Sequential Simulation: the Concept
• Steps in Sequential Simulation
• SISIM Program

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Sequential Simulation: the
Concept
2

1
3
1. Assign data values to closest grid node
2. Establish a random path through all of the grid nodes
3. Visit each grid node:
(a) find nearby data and previously simulated grid nodes
(b) construct the conditional distribution by kriging (this is where the
variogram comes in)
(c) draw simulated value from conditional distribution
4. Check the results

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
STEP 1

Assign Data to Grid Nodes


Data

Data
Data

Data
Why?
• Explicitly honor data ⇒ data values will appear in final 3-D model
• Improves the CPU speed of the algorithm: searching for previously simulated
nodes and original data is accomplished in one step
Considerations:
• Take the closest of multiple data within the same cell ⇒ could lose some
information (see in figure )
• Not an option when simulating a cross-sectional or small-area model

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
STEP 2

Establish a Random Path


32 4 34 18 35 2 21
30 31 14 8 9 16 3
28 6 24 5 15 13 18
11 5 28 20 27 23 10
1 27 19 33 25 4 26
29 29 7 12 17 6 22
• Visit each cell once and only once in random order
• Can do this in many ways:
– draw a random number and multiply it by N
– sort an array of random numbers while carrying an array of indices
capitalize on the limited period length of linear congruential generators
• Skip over cells (actually grid nodes) that already have a value

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
STEP 3 - (a)

Find Nearby “Informed” Nodes

• “Informed” nodes refers to both data-nodes and nodes that have been informed
earlier in the random path
• Typically use spiral search to identify the close nodes
• Limit the number of nodes actually considered:
– octant search (?)
– maximum per octant (say 4)
– maximum number

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
STEP 3 - (b)
Construct Conditional
Distribution

• Conditional distribution is constrained by:


– global proportion of each lithology type
– local data
– “local” proportion from secondary data such as
seismic
• Calculate by kriging the binary indicator transform
for each rock type

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Indicator Simulation (1)
• Define an indicator transform:

i (uα ; k ) = 1, i f lithosfacie s k
0, i f not
present a t locatio n uα

• Average of an indicator is the global proportion:


proportion of k = E { I ( uα ; k )}
n

∑w
α =1
α ⋅ i (uα ;k )
=
n

wα values account for data clustering

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Indicator Simulation (2)
• The variogram of an indicator variable measures spatial correlation:
1
γ I ( h) = E {[ I ( u; k ) − I ( u + h; k )]2 }
2
Vertical Distance, Zrel
0 1.0
0.25

Vertical Variogram

γ
Horizontal Variogram

0.00
0 Horizontal Distance, m 1600

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
STEP 3 - (c)
Construct Conditional
Distribution with Kriging

• Given n nearby data values k(ui),i=1,...,n how do we calculate the conditional


distribution?
• Estimate conditional probabilities of each rock type pk*(u), k=1,...,K by a
linear combination of the nearby data:
n n
p ( u) = ∑ λα ( u) ⋅ I ( uα ; k ) + [1 − ∑ λα ( u)] ⋅ mk
*
k
α =1 α =1

• Determine weights λα(u), α=1,...,n by the well known “normal system” or


kriging.
• Kriging weights account for two things:
– clustering of the data locations
– closeness of the data to the location being considered
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
STEP 3 - (d)
Draw a Simulated Value
1

...

0
1 2 3
• probabilities p*k(u), k=1,...,K are given by kriging
• Procedure:
– draw a random number ∈[0,1]
– find which class k is specified by the random number
– assign k to node
• Since the conditional probabilities were estimated by kriging with a given
variogram γk(h), k=1,...,K, the simulated values, taken all together, will
reproduce those variograms ,γk(h), k=1,...,K

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Detailed Steps in SISIM
1. Establish grid network and coordinate system (Zrel-space)
2. Assign data to the nearest grid node (take the closest of multiple data assigned
to same node)
3. Determine a random path through all of the grid nodes
(a) find nearby data and previously simulated grid nodes
(b) construct the conditional probabilities by kriging
(c) draw simulated value from conditional distribution
4. Check results
(a) honor data?
(b) honor global proportions?
(c) honor variogram?
(d) look reasonable

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
G
eo
s t
Parameters for SISIM a
va tist
A
******************** G
ila ic SL
bl al
START OF PARAMETERS: ea So IB
1 \ 1= continuous(cdf), 0= ca teg orical(pdf) t w ftw
w a
w
5 \ numb er thres hold s /categ ories .G re
0.5 1.0 2.5 5.0 10.0 \ thres hold s / ca teg ories LS LI
IB Br
0.12 0.29 0.50 0.74 0.88 \ glob a l cd f / p d f .co ar
../ data/ clus ter.dat \ file w ith data m y
1 2 0 3 \ columns for X,Y,Z, and variab le
d irect.ik \ file w ith s oft ind icator inp ut
1 2 0 34567 \ columns for X,Y,Z, and indicators
0 \ Markov-Bayes simulation (0= no,1= yes )
0.61 0.54 0.56 0.53 0.29 \ ca lib ra tion B(z) va lues
-1.0e21 1.0e21 \ trimming limits
0.0 30.0 \ minimum and maximum d ata va lue
1 0.0 \ low er tail option and parameter
1 1.0 \ middle option and parameter
1 30.0 \ upper tail option and parameter
clus ter.d at \ file w ith tab ulated values
3 0 \ columns for variable, w eig ht
0 \ deb ug g ing level: 0,1,2,3
s is im.dbg \ file for debugging output
s isim.out \ file for s imulation outp ut

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
G
eo
s t
A at
1 \ numb er of realizations v ai
ist G
la ic SL
bl al
50 0.5 1.0 \ nx,xmn,xs iz ea So IB
50 0.5 1.0 \ ny,ymn,ysiz tw ftw
1 1.0 10.0 \ nz,zmn,zs iz w ar
w eL
69069 \ rand om numb er seed .G
LS IB
12 \ maximum original d ata for each kriging IB
12 \ maximum p revious nod es for each krig ing .co rar
m y
1 \ maximum s oft ind icator nod es for kriging
0 \ a ssig n da ta to nod es ? (0= no,1= yes )
0 3 \ multip le g rid s earch? (0= no,1= yes ),num
0 \ maximum p er octant (0= not us ed )
20.0 20.0 20.0 \ ma ximum s ea rch ra dii
0.0 0.0 0.0 \ a ng les for s earch ellip s oid
0 2.5 \ 0= full IK, 1= med ian ap p rox. (cutoff)
0 \ 0= SK, 1= OK
1 0.15 \ One ns t, nug g et effect
1 0.85 0.0 0.0 0.0 \ it,cc,ang 1,a ng2,ang 3
10.0 10.0 10.0 \ a _hma x, a_hmin, a _vert
1 0.10 \ Tw o ns t, nug g et effect
1 0.90 0.0 0.0 0.0 \ it,cc,ang 1,a ng2,ang 3
10.0 10.0 10.0 \ a _hma x, a_hmin, a _vert
1 0.10 \ Three ns t, nug g et effect
1 0.90 0.0 0.0 0.0 \ it,cc,ang 1,a ng2,ang 3
10.0 10.0 10.0 \ a _hma x, a_hmin, a _vert

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Reservoir Modeling with GSLIB

Gaussian Simulation for Porosity


Modeling
• Petrophysical Property Simulation
• Gaussian Simulation
• Sequential Gaussian Simulation
• More Comments on the Steps in Sequential Simulation
• SGSIM Program

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Petrophysical Property
Modeling: Prerequisites
• Work within “homogeneous” lithofacies/rock-type classification a may
require a first step to model lithology
• Sequence stratigraphic framework a Zrel vertical coordinate space
• Clean data: positioned correctly, manageable outliers, grid spacing is
appropriate
• Need to understand special features and “special” data:
– trends
– production data
– seismic data
• Considerations for areal grid size:
– practical limit to the number of cells
– need to have sufficient resolution so that the upscaling is meaningful
– this resolution is required even when the wells are widely spaced
(simulation algorithms fill in the heterogeneity)
• Work with “grid nodes”. We assign a property for the entire cell knowing that
there are “sub-cell” features

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Different Petrophysical
Property Simulation
Algorithms
• Matrix Approach (LU Decomposition): not used because of the size of the
problem (an N x N matrix must be solved where N could be in the millions)
lus im
• Turning Bands: simulate the variable on 1-D lines and combine in 3-D. Not
used because of artifacts and difficulty in conditioning to local data tb3d
• Fractals: not extensively used because self-similarity is only valid for a
limited range of scales and difficulty in conditioning to local data (see Tom
Hewett)
• Annealing: becoming popular a recommended for permeability (talk about
later)
• Sequential Simulation: widely used and recommended

⇒ main purpose is to describe Sequential Gaussian Simulation (SGSIM)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Estimation versus Simulation
True Distribution Simulation

Kriging

Location Map of Sample Data Multiple Simulated Realizations

• Estimation is locally accurate and smooth, appropriate for visualizing trends,


inappropriate for flow simulation where extreme values are important, and does not
assess of global uncertainty
• Simulation reproduces histogram, honors spatial variability (variogram), a appropriate
for flow simulation, allows an assessment of uncertainty with alternative realizations
possible
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Sequential Simulation: Theory
n
• Recall the kriging estimator: Y * ( u) = ∑ λβ ⋅ Y ( uβ )
β =1

and the corresponding kriging system: n

∑ λ C (u
β =1
β α
, uβ ) = C ( u, uα ) , Vuα

• The kriging system forces the covariance between the kriged estimate and the data
values to be correct: n
Cov {Y ( u ),Y ( uα )} = ∑ λ β C ( uα , uβ ) = C ( u, uα ) = C {Y ( u),Y ( uα )}
*

β =1

• Although the covariance between the estimates and the data is correct, the variance is
too small:
Var {Y * ( u )} = C (0) − σ SK
2
( u)
correct the variance without changing the covariance by adding an independent
(random) component with the correct variance:
YS ( u ) = Y * ( u ) + R ( u )
where R(u}) corrects for the missing variance.
• Covariance between kriged/simulated values is not correct:
Cov{Y * ( u ),Y * ( u' )} ≠ C {Y ( u ),Y ( u' )}
• The idea of sequential simulation is to use previously kriged/simulated values as data a
reproduce the covariance between all of the simulated values!

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Why Sequential Gaussian
Simulation? (1)
Gaussian distribution is used because it is extraordinarily straightforward to
establish conditional distributions: shape of all conditional distributions is
Gaussian (normal) and the mean and variance are given by kriging
1. Transform data to normal scores in the beginning (before variography)
2. Simulate 3-D realization in “normal space”
3. Conditional distributions are calculated by kriging to honor:
– global histogram: N(0,1)
– local data
– secondary data (seismic, production data, ...)
4. Back-transform all of the values when finished
⇒ Price of mathematical simplicity is the characteristic of maximum spatial
entropy, i.e.,
low and high values are disconnected. Not appropriate for permeability.

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Why Sequential Gaussian
Simulation? (2)
Steps in SGSIM:
1. Transform data to “normal space”
2. Establish grid network and coordinate system (Zrel-space)
3. Decide whether to assign data to the nearest grid node or keep separate
4. Determine a random path through all of the grid nodes
(a) search for nearby data and previously simulated grid nodes
(b) construct the conditional distribution by kriging
(c) draw simulated value from conditional distribution
5. Back transform and check results

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Normal Scores Transformation
“graphical” one-to-one (rank preserving) transform

Cumulative Distribution Cumulative Distribution


Cumulative Frequency

Porosity Normal Scores


Frequency

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Transformation to/from
Normal Space
• Declustering weights are necessary
• Use “global” transformation table for small-area or cross-sectional models a
there may be too few data in the area of interest
• “Tail” options:
– typically simulate the property at many more grid nodes than there are
data a expect higher values than observed in the data (and lower).
– almost always a linear interpolation to user-defined minimum and
maximum values is acceptable
– can account for skewness in the upper tail with a hyperbolic model

5.0

1.0
λ
Fω ,λ ( z ) = 1 − , ω ≥1

F(zk)

zk Z variable
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Establish Data and Grid Network
• Work within “homogeneous” lithofacies/rock-type classification a may
require a first step to model lithology
• Zrel vertical coordinate space
• Clean data: positioned correctly, manageable outliers, grid spacing is
appropriate
• Considerations for areal grid size:
– practical limit to the number of cells
– need to have sufficient resolution so that the upscaling is meaningful
– this resolution is required even when the wells are widely spaced
(simulation algorithms fill in the heterogeneity)
• Work with “grid nodes”. We assign a property for the entire cell knowing that
there are “sub-cell” features

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Search: Two-Part or Assign
Data to Grid Nodes

Two-Part?
• search for previously simulated nodes and then original data in two steps a then treat
the same when constructing the conditional distribution
• honor the data at their locations even if they can not be seen in the final model
• necessary for cross-sectional or small-area models
Assign Data to Grid Nodes:
• explicitly honor data - data values will appear in final 3-D model
• improves the CPU speed of the algorithm: searching for previously simulated nodes and
original data is accomplished in one step

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Multiple Grid Concept

• We typically limit ourselves to the nearest 12-48 grid nodes


• As the sequential simulation proceeds (fine grid) only the close samples will be used
• Variogram reproduction can be poor, especially if the variogram range is large with
respect to the grid node spacing
• Multiple grid concept: simulate a coarse grid first and then refine one or more times
• Searching for data is more complicated:
– two-part search at coarse levels
– perhaps relocate to nearest fine grid node first (unless a two-part is used at the finest level)
• A number of implementation considerations

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Number of Data to Consider
Reasons for more:
• theoretically better
• more accurate estimate of the conditional mean and variance
• better reproduction of the variogram
Reasons for less:
• CPU time is proportional to N3
• memory requirements proportional to N2
• negative weights are commonly encountered when data are screened
• using fewer data places less emphasis on the assumption of stationarity
So, choose between 12 to 48 depending on:
• 2-D versus 3-D
• range of variogram relative to grid node spacing
• CPU time restrictions

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Type of Kriging
• Simple Kriging (SK):
m *SK = ∑ λi ⋅ z ( ui ) + 1 − ∑ λi  ⋅ m global
n n

i =1
 i =1 

• Ordinary Kriging (OK) - constrain


n
*
mOK = ∑ λi ⋅ z ( ui )
i =1

• Other Types:
– Universal Kriging (UK) a accounts for simple trends
– External Drift a accounts for more complex trends
– Locally Varying Mean a accounts for secondary information

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Detailed Steps in SGSIM
1. Transform data to “normal space”
2. Establish grid network and coordinate system (Zrel-space)
3. Assign data to the nearest grid node (take the closest of multiple data assigned
to the same node)
4. Determine a random path through all of the grid nodes
(a) find nearby data and previously simulated grid nodes
(b) construct the conditional distribution by kriging
(c) draw simulated value from conditional distribution
5. Check results
(a) honor data?
(b) honor histogram: N(0,1) - standard normal with a mean of zero and a
variance of one?
(c) honor variogram?
(d) honor concept of geology?
6. Back transform

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
G
eo
s t
a
va tist
A
G
ila ic SL
al
SGSIM Program (1) bl
ea
tw
w
w
So IB
ftw
a
Parameters for SGSIM .G re
LS LI
******************** IB B
.co rar
START OF PARAMETERS: m y
../data/cluster.dat \ file with data
1 2 0 3 5 0 \ columns for X,Y,Z,vr,wt,sec.var.
-1.0 1.0e21 \ trimming limits
1 \ transform the data (0=no, 1=yes)
sgsim.trn \ file for output trans table
0 \ consider ref. dist (0=no, 1=yes)
histsmth.out \ file with ref. dist distribution
1 2 \ columns for vr and wt
0.0 15.0 \ zmin,zmax(tail extrapolation)
1 0.0 \ lower tail option, parameter
1 15.0 \ upper tail option, parameter
1 \ debugging level: 0,1,2,3
sgsim.dbg \ file for debugging output
sgsim.out
5
\ file for simulation output
\ number of realizations to generate
...
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
G
eo
s t
a
va tist
A
G
ila ic SL
bl al
ea So IB
SGSIM Program (2) tw
w
w
ftw
.G re
a
LS LI
50 0.5 1.0 \ nx,xmn,xsiz IB B
50 0.5 1.0 \ ny,ymn,ysiz .co rar
m y
1 0.5 1.0 \ nz,zmn,zsiz
69069 \ random number seed
0 8 \ min and max original data for sim
12 \ number of simulated nodes to use
1 \ assign data to nodes (0=no, 1=yes)
1 3 \ multiple grid search (0=no, 1=yes),num
0 \ maximum data per octant (0=not used)
10.0 10.0 10.0 \ maximum search radii (hmax,hmin,vert)
0.0 0.0 0.0 \ angles for search ellipsoid
4 0.60 \ ktype: 0=SK,1=OK,2=LVM,3=EXDR,4=COLC
../data/ydata.dat \ file with LVM, EXDR, or COLC variable
4 \ column for secondary variable
1 0.1 \ nst, nugget effect
1 0.9 0.0 0.0 0.0 \ it,cc,ang1,ang2,ang3
10.0 10.0 10.0 \ a_hmax, a_hmin, a_vert

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Reservoir Modeling with GSLIB

Case Studies / Modeling Tips


• Sequential Approach to Reservoir Modeling
• Question / Answer Time
• A Small Example
• Glimpses of Case Studies

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Reservoir Modeling
Establish Stratigraphic Layering / Coordinates

Cell-based Lithofacies Object-Based Lithofacies


Modeling Modeling
Repeat for Multiple Realizations

Porosity Modeling
Model Uses
1. Volumetric / Mapping
2. Assess Connectivity
3. Scale-Up for Flow Simulation
4. Place Wells / Process Design
Permeability Modeling

Main geostatistical modeling flow chart: the structure and stratigraphy of each reservoir
layer must be established, the lithofacies modeled within each layer, and then porosity and
permeability modeled within each lithofacies.

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Introductory Example
Existing Vertical Well
Top of Reservoir

• Fashioned after a real problem and the geological data is based on outcrop observations
• A horizontal well is to be drilled from a vertical well to produce from a relatively thin oil
column.
• The goal is to construct a numerical model of porosity and permeability to predict the
performance of horizontal well including (1) oil production, (2) gas coning, and (3)
water coning.

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Introductory Example -
Petrophysical Data
•Lithofacies 1
10,000 •Lithofacies 2
•Lithofacies 3
Permeability, md

Code Lithofacies Average Coefficient K v :K h


Perm. of variation ratio
0 Coal and Shale 1 md 0.00 0.1
1 Incised Valley Fill Sandstone 1500 md 1.00 1.0
2 Channel Fill Sandstone 500 md 1.50 0.1
3 Lower Shoreface Sandstone 1000 md 0.75 0.8

10
0 0.4
Porosity

Permeability characteristics of each lithofacies: the coefficient of variation is the


average permeability divided by the standard deviation, Kv is the vertical permeability,
and Kh is the horizontal permeability.

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Flow Simulation

Gridding for flow simulation. For numerical efficiency, the vertical gridding is
aligned with the gas-oil fluid contact and the oil-water fluid contact. The black
dots illustrate the location of the proposed horizontal well completions.
Representative three-phase fluid properties and rock properties such as
compressibility have been considered. It would be possible to consider these
properties as unknown and build that uncertainty into modeling; however, in this
introductory example they have been fixed with no uncertainty.

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Simple Geologic Models
Layercake Model Gaussian Simulation Model

Smooth Model

Three simple assignments of rock properties (a) a “layercake” or horizontal


projection model, (b) a smooth inverse distance model, and (c) a simple Gaussian
simulation.
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Simple Geologic Models:
Flow Results
1000
Oil Production Ratio (m3/day)

1.0
1600

Gas Oil Ratio


Water Cut

0 3000 0 3000 0 3000


Time (days) Time (days) Time (days)

Flow results: layercake model - solid line; smooth model - long dashes; simple
geostats model -- short dashes.

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Better Geologic Model
(a) Geostatistical Model (b) Geostatistical Model - Flow Grid

The first geostatistical realization shown on the geological grid and the flow
simulation grid

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Multiple Realizations

01 06 11 16

02 07 12 17

03 08 13 18

04 09 14 19

05 10 15 20

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Geologic Models - Flow Results
10,000 1.0 2000

Gas Oil Ratio


Oil Production

Water Cut
m3/day

0 3000 0 0 3000
3000
Time, days Time, days Time, days

Flow results from 20 geostatistical realizations (solid gray lines) with simple
model results superimposed

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Uncertainty
Cumulative Oil Production at 1000 Days Time of Water Break Through

layercake first sgsim


first sgsim smooth
smooth
layercake

Frequency
Frequency

200 800 0 1600

Cumulative Oil Production, x 10^3 m^3 Water Break Through, days

The cumulative oil production after 1000 days and the time to water breakthrough.
Note the axis on the two plots. There is a significant difference between the
simple models and the results of geostatistical modeling (the histograms).

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Major Arabian Carbonate
Reservoir
• GOSP 2 & 7 Area study commissioned by Zone Porosity % Layer
Saudi Aramco
• SPE29869 paper Integrated Reservoir
Modeling of a Major Arabian Carbonate 3A
Reservoir by J.P. Benkendorfer, C.V. Deutsch,
P.D. LaCroix, L.H. Landis, Y.A. Al-Askar,
A.A. Al-AbdulKarim, and J. Cole
• Oil production from wells on a one-kilometer 3B
spacing with flank water injection. There has
been significant production and injection
during the last 20 years 3A
• This has had rapid and erratic water movement
uncharacteristic of the rest of the field a
reason for building a new geological and flow 3B
simulation models

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Modeling Process
Matrix Large-Scale
Lithology Porosity
Permeability Permeability

• Standard GSLIB software (because it was for Saudi Aramco)


• Novel aspect was modeling permeability as the sum of a matrix permeability
and a large-scale permeability
– fractures
– vuggy and leached zones
– bias due to core recovery
• Typical modeling procedure that could be applied to other carbonates and to
clastic reservoirs

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Indicator Simulation of
Lithology
Vertical Indicator Variogram: Layer 8 Horizontal Indicator Variogram: Layer 8
120 degrees
0.25 0.25
30 degrees

γ γ

0 0.8 0 100
Distance Distance
Limestone N

Dolomite

1 km

Presence / absence of limestone / dolomite was modeled with indicator simulation


(SISIM) on a by-layer basis

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Gaussian Simulation of

Porosity
Variogram model for porosity in limestone:
Vertical Porosity Variogram Horizontal Porosity Variogram
Layer 8 (Limestone) Layer 8 (Limestone)

1.0 110 degrees


0.8
Variogram

Variogram
20 degrees

0 0
0 0.8 0 10,000
Distance (m) Distance (m)
• Variogram model for porosity in dolomite:
Vertical Porosity Variogram Horizontal Porosity Variogram
Layer 8 (Limestone) Layer 8 (Limestone)
110 degrees
0.8 Variogram 1.0
Variogram

20 degrees

0 0
0 0.8 0 10,000
Distance (m) Distance (m)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Gaussian Simulation of
Porosity

Porosity models for limestone and dolomite were built on a by-layer basis with
SGSIM and then put together according to the layer and lithology template

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Indicator Simulation of
Matrix Permeability
Group 1 - Limestone Vertical Matrix k Variogram Layer 5 (Limestone)
2.0
Means: Correlation: Linear Transform:
Porosity = 21.57 Pearson = 0.73 Slope = 0.129

Variogram
Permeability = 295.7 Spearman = 0.70 Intercept = -1.224
10,000
Permeability (md)

0
0 1.0
Stratigraphic Distance (m)
Horizontall Matrix k Variogram Layer 5 (Limestone)
2.0 20 degrees
110 degrees

0.01 Variogram
0
40
Porosity (%) 0
0 12000
Stratigraphic Distance (m)
Numbers above x-axis are porosity class percentages
Numbers at corners are porosity/permeability class percentages

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Gaussian Simulation of
Large-Scale Permeability
Vertical Large Scale Variogram Layer 5 Horizontal Large Scale Variogram Layer 5
2.0 2.0
isotropic

Variogram
Variogram

0 0
0 1.0 0 12000
Stratigraphic Distance Stratigraphic Distance

• Matrix permeability at each well location yields a K•hmatrix


• Well test-derived permeability at each well location yields a K•htotal
• Subtraction yields a K•hlarge
• Vertical distribution of K•hlarge scale on a foot-by-foot basis is done by
considering multiple CFM data

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Gaussian Simulation of
Large-Scale Permeability

• Large-scale permeability models were built on a by-layer basis with SGSIM


• Matrix permeability and large-scale permeability models were added together
to yield a geological model of permeability
• A calibrated power average was considered to scale the geological model to
the resolution for flow simulation

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Flow Simulation: First
History Match
3400
Datum Pressure (psi)

1600
100
Water Cut (%)

0
1975 1980 1985 1990 1994
1940 1995
Year

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Flow Simulation: Fourth
History Match
3400
Datum Pressure (psi)

1600
100
Water Cut (%)

0
1975 1980 1985 1990 1994
1940 1995
Year

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Reservoir Modeling with GSLIB

Indicator Simulation for


Permeability Modeling

• Discuss the Problem of Accounting for Secondary Data


• Review Conventional Techniques
• Sequential Indicator Simulation
• Examples

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Indicator Coding of Data
• local hard indicator data i(uα;z) originating from local hard data z(uα):
≤z, = 0 if not
i(uα;z) = 1 if z(uα )≤
• local hard indicator data j(uα;z) originating from ancillary information that
∈(a α,b α],
provides hard inequality constraints on the local value z(uα). If z(uα)∈
then:
0 if z ≤ a α
j(uα;z) = undefined (missing) if z∈(a α,b α]
1 if z > b α
• local soft indicator data y(uα;z) originating from ancillary information providing
prior (pre-posterior) probabilities about the value z(uα ):
y(uα;z) = Prob {Z(uα) ≤ z | local information}
∈[0,1], and ≠ F(z) : global prior as defined hereafter
• global prior information common to all locations u within the stationary area A:
F(z) = Prob{Z(u) ≤ z}, ∀ u ∈ A

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Indicator Kriging (IK)
Example Data
1.0

Cumulative Frequency
Frequency

0
0 12.0 0 12.0
Variable
• Build a local cdf conditional to surrounding data
• Can compute any of the following:
– E-type estimate (local conditional mean)
– Probability to exceed a threshold z
– z-value of any probability
– probability intervals
– truncated statistics

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Markov Bayes Model (1)
The IK process can be seen as a Bayesian updating of the local prior cdf into a
posterior cdf using information supplied by neighboring local prior cdf's
n n'
[Pr ob{ Z ( u) ≤ z | ( n + n' )}]*IK = λ0 ( u )F ( z ) + ∑ λα ( u; z )i ( uα ; z ) + ∑υα ' ( u; z ) y( u'α ; z )
α =1 α ' =1

The λα(u;z)’s are the weights attached to the n neighboring hard indicator data, the
να(u;z)’s
are the weights attached to the n´ neighboring soft indicator data, and λo is the
weight attributed to the global prior cdf. To ensure unbiasedness, λo is usually set
to:
n n'
λ ( u) = 1 − ∑ λ ( u; z ) − ∑υ ( u; z )
0 α α'
α =1 α '= 1

This can be seen as an indicator cokriging that pools information of different


types: the hard i and j indicator data and the soft y-prior probabilities.

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Markov Bayes Model (2)
The Markov-Bayes model is a model whereby the matrix of covariances is given
by the following model.
C IY ( h; z ) = B( z )C I ( h; z ) , V h
CY ( h; z ) = B 2 ( z )C I ( h; z ) , V h > 0
= B( z ) C I ( h; z ) , h = 0

The coefficients B(z) are obtained from calibration of the soft y-data to the hard z-
data; more precisely:

B ( z ) = m (1 ) ( z ) − m ( 0 ) ( z ) ∈ [−1,+1]

with:
m ( 1) ( z ) = E { y( u; z ) | I ( u; z ) = 1}
m ( 0 ) ( z ) = E { y( u; z ) | I ( u; z ) = 0}

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Order Relations Corrections

Order Relations Correction Order Relations Correction


(no data in class z6 and z7)
1 1

0 0
Zmin z1 z2 z3 z4 z5 z6 z7 z8 zmax Zmin z1 z2 z3 z4 z5 z6 z7 z8 zmax

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
G
eo
s t
a
va tist
A
G
ila ic SL
bl al
ea So IB
Parameter File (1) tw
w
w
ftw
.G re
a
LS LI
IB B
.co rar
Parameters for SISIM m y
********************
START OF PARAMETERS:
1 \ 1=continuous(cdf), 0=categorical(pdf)
5 \ number thresholds/categories
0.5 1.0 2.5 5.0 10.0 \ thresholds / categories
0.12 0.29 0.50 0.74 0.88 \ global cdf / pdf
../data/cluster.dat \ file with data
1 2 0 3 \ columns for X,Y,Z, and variable
direct.ik \ file with soft indicator input
1 2 0 34567 \ columns for X,Y,Z, and indicators
0 \ Markov-Bayes simulation (0=no,1=yes)
0.61 0.54 0.56 0.53 0.29 \ calibration B(z) values
-1.0e21 1.0e21 \ trimming limits
0.0 30.0 \ minimum and maximum data value ...
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
G
eo
s t
a
va tist
A
G
ila ic SL
bl al
ea So IB
Parameter File (2) tw
w
w
ftw
.G re
a
LS LI
IB B
.co rar
1 0.0 \ lower tail option and parameter m y
1 1.0 \ middle option and parameter
1 30.0 \ upper tail option and parameter
cluster.dat \ file with tabulated values
3 0 \ columns for variable, weight
0 \ debugging level: 0,1,2,3
sisim.dbg \ file for debugging output
sisim.out \ file for simulation output
1 \ number of realizations
50 0.5 1.0 \ nx,xmn,xsiz
50 0.5 1.0 \ ny,ymn,ysiz
1 1.0 10.0 \ nz,zmn,zsiz
69069 \ random number seed
12 \ maximum original data for each kriging
12 \ maximum previous nodes for each kriging ...
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
G
eo
s t
a
va tist
A
G
ila ic SL
bl al
ea So IB
Parameter File (3) tw
w
w
ftw
.G re
a
LS LI
IB B
.co rar
12 \ maximum previous nodes for each kriging m y
1 \ maximum soft indicator nodes for kriging
1 \ assign data to nodes? (0=no,1=yes)
0 3 \ multiple grid search? (0=no,1=yes),num
0 \ maximum per octant (0=not used)
20.0 20.0 20.0 \ maximum search radii
0.0 0.0 0.0 \ angles for search ellipsoid
0 2.5 \ 0=full IK, 1=median approx. (cutoff)
0 \ 0=SK, 1=OK
1 0.15 \ One nst, nugget effect
1 0.85 0.0 0.0 0.0 \ it,cc,ang1,ang2,ang3
10.0 10.0 10.0 \ a_hmax, a_hmin, a_vert
1 0.10 \ Two nst, nugget effecT

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Reservoir Modeling with GSLIB

Annealing Techniques for Data


Integration
• Discuss the Problem of Permeability Prediction
• Present Annealing Cosimulation
• More Details on Simulated Annealing
• Examples
• SASIM program

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Regression-Type Deterministic
Approaches
1000 Calibration Data

Approaches: Deterministic
Answer

Permeability (md)
• linear regression - 105 md
• quadratic, cubic, ... regression
• porosity class average or conditional averages
Characteristics:
• smooth out low and high values
Known Porosity Value
• does not capture uncertainty 1
0.0 30.0
• transformed permeability have incorrect spatial variability Porosity (%)
Considerations:
• extreme high and low values have the most impact on fluid flow
• spatial correlation (connectivity) is very important fewer “hard” permeability
K data than porosity φ
• K is correlated with porosity φ
• build φ model first then K model (exhaustive secondary variable)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Stochastic Approaches
1000 Calibration Data
Conditional
Deterministic Distribution
Answer
Permeability (md)

- 105 md

Known Porosity Value


1
0.0 30.0
Porosity (%)
Approaches:
• stochastic simulation from porosity classes
• Markov-Bayes (implemented as a sequential simulation algorithm)
• collocated cosimulation (Gaussian)
Characteristics:
• can extract a single expected value (for looking at trends)
• calculate probability intervals (90% interval: 44-210 md)
• ⇒ DRAW SIMULATED VALUES

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Annealing Techniques to Account
for a Secondary Variable
What is Annealing?
Annealing, or more properly simulated annealing, is an optimization algorithm
based on an
analogy with the physical process of annealing.
• Treat O as an energy function
• Cool an initial realization:
– perturb system to simulate thermal agitation
– always accept swaps that lower O
– sometimes accept swaps that increase O
– cool slowly ⇒ find a minimum energy solution
What is Cosimulation?
Cosimulation is the act of generating a numerical model of one variable that is
conditional to
the results of another variable, for example,
• model permeability conditional to porosity
• model porosity conditional to log data
• simulate multiple variables sequentially

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Simulated Annealing
• Simulated annealing is a solution method in the field of combinatorial
optimization based on an analogy with the physical process of annealing.
Solving a combinatorial optimization problem amounts to finding the ‘best’ or
‘optimal’ solution among a finite or countable infinite number of alternative
solutions.
• Introduced in the early 1980's by Kirkpatrick, Gelatt & Vecchi [1992;1983]
and independently Cerny [1985]
• “Simulating the annealing process” dates back to 1953 and the work of
Metropolis, Rosenbluth, Rosenbluth, Teller & Teller
• Applications in Spatial Statistics:
– Geman and Geman, 1984
– Farmer, 1989
– Others, 1990-present
• In the application of annealing there is no explicit random function model,
rather, the creation of a simulated realization is formulated as an optimization
problem to be solved with a stochastic relaxation or “annealing” technique.

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Application of Simulated
Annealing
Prerequisites to apply simulated annealing as a numerical optimization technique:
• description of the system
• quantitative objective (energy) function
• random generator of moves or rearrangements
• an annealing schedule of the temperatures and the lengths of time to let the
system evolve at each temperature
Some example applications:
• studying the behavior of materials such as crystals, magnetic alloys, and spin
glasses
• travelling salesman-type problems
• routing of garbage collection trucks
• wiring layout of computers and circuit layout on computer chips
• assisting with seismic inversion
• geostatistics

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Steps in Annealing-Based
Simulation
1. Establish an initial guess that honors the well data
Assign a K value to each cell by drawing from the conditional distribution of K
given the cell’s φ
2. Calculate the initial objective function
Numerical measure of mismatch between the desired variogram and the one of
the initial guess
3. Consider a change to a cell’s permeability
Randomly choose a non-data cell and then consider a new K from the
conditional distribution of K given the cell’s φ
4. Evaluate new objective function
• better? - accept change
• worse? - reject change
5. Is objective function close enough to zero?
• yes - done
• no - go to 3

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
An Example
nh
O = ∑ [γ Z* ( hi ) − γ Z ( hi )]2
i =1

Starting Image Half Way Final Distribution

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
The Objective Function
Honor the Porosity/Permeability Cross-plot:
10,000
nφ nk
Oc = ∑ ∑ [ f (φ i , K j )calibration − f (φ i , K j )realization ]2

Permeability
i =0 j =0

Honor the Variogram: 0.01


0.0 0.2
Permeability Variogram Model Porosity

2.0
Variogram

n
Oc = ∑ [γ imod el − γ irealization ]2
i =1

0.0
0.0 40.0
Distance (m)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Porosity Profile
A Simple Example
35
0

Calibration Data Permeability Semivariogram Model

3.0
2.0

Variogram
Log10 KH

1.0

-2.0 0.0
0.0 40.0 0.0 40.0
Porosity Distance (m)

• Generate corresponding permeability values


• Need an objective function for ACS

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
A Simple Example: Results
Porosity Profile Permeability Profile

-2.0

3.0
35
0

Calibration Data
Down Well Semivariogram:
model and simulated
3.0

Variogram
2.0
Log10 KH

1.0

0.0
0.0 40.0
-2.0 Distance (m)
0.0
Porosity
40.0

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Weighting Different Constraints
• The weights νc allow equalizing the contributions of each component in the
n
global objective function O = ∑ν ⋅ O c c
c =1

• Each weight νc is made inversely proportional to the average change in


absolute value of its component objective function:
1
νc = , c = 1,..., C
∆Oc
• ∆Oc is numerically approximated by:
1 M (m)
∆Oc = ∑
M m =1
| Oc − Oc | , c = 1,..., C

• The overall objective function may then be written as,


1.0
Object Function Values

mean
1 C variance

(0) ∑ c
O= ⋅ ν ⋅ Oc smoothness
O c =1
quantiles

0.0
0.0 50,000
Number of Perturbance
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Scale and Precision of
Seismic Data

Φ 30-70 ft.

75-150 ft.

Seismic Attributes

• Geological models have greater vertical resolution than that provided by


seismic data (areal resolution is comparable)
• Seismic attribute (impedance, integrated energy, ...) does not precisely inform
the vertical average of porosity
• May also imprecisely inform the relative proportion of specific rock types
• Very valuable information due to the near exhaustive coverage

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Annealing Approach
• Consider the annealing procedure:
1. Establish an initial realization that honors the well data
2. Calculate the initial objective function
3. Consider a change to a cell's permeability
4. Evaluate new objective function (better? - accept change; worse? - reject
change)
5. Is objective function close enough to zero? (yes - done; no - go to 3)
• Add component objective function(s) that capture the correlation with between
vertical averages of the porosity (rock type proportion) and the seismic
attribute
nφ nk
Oc = ∑ ∑ [ f (φ i , K j )calibration − f (φ i , K j )realization ]2
i =0 j =0

Oc = [ ρ calibration − ρ realization ]2
• Where ρ is defined between the vertically averaged porosity and the seismic
attribute, S is the seismic average, and P is the vertically averaged porosity

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Application from West Texas

• West Texas Permian Basin (data provided to SCRF for technique development)
• 74 wells in the area (50 within the area covered by the 3-D seismic survey)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Seismic Attribute Data
Seismic Attribute
130
25000

20000

15000
North

10000

5000

0.0

0.0
0.0 East 130

• 130 by 130 - 80 foot square areal pixels


• Significant areal variation

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Seismic Attribute Data
0.06

Frequency

0.00
0. 25000
Seismic Attribute
• spike of zero values
• relatively low coefficient of variation

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
13.0
Calibration Data

Porosity

ρ = 0.54

3.0
0 25000
Seismic Attribute
• Positive correlation between the vertically averaged porosity and seismic
attribute
• Linear correlation coefficient of 0.54 is typical
• Calibration covers the range of seismic values (this can be a significant
problem when there are few wells)

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Porosity Histogram Porosity
0.10

Frequency

0.0
0.0 25.0
Porosity
• Greater variance than 2-D vertical average (as expected)
• Declustering and perhaps smoothing should be considered to get representative
histogram
• 3-D models will, within ergodic fluctuations, replicate this histogram.
Consider a resampling procedure to assess uncertainty in porosity histogram

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Porosity Variogram
Vertical Variogram Horizontal Variogram

1.0 1.0

Variogram
Variogram

type sill range type sill range


spherical 0.4 1.1 spherical 0.4 500
spherical 0.6 15 spherical 0.6 4000

0.0 0.0
0.0 20.0 0.0 20.0
Distance Distance

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
3-D Porosity Modeling
• Annealing-based simulation constrained to:
– local well data
– 99 evenly spaced quantiles of the porosity histogram
– 50 variogram lags
– correlation coefficient of 0.54 between the vertically averaged porosity
and the seismic attribute
• Can create multiple realizations

Vertical Average of Porosity from Model


• Show one for illustration
• Cross plot from model

Seismic Energy

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Horizontal Slices
Slice 30 (near top) Slice 10 (near bottom)
130 130 20.0

15.0
North

North
10.0

5.00

0.0
0.0 0.0
0.0 East 130 0.0 East 130

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Vertical Average

Vertical Average Seismic Attribute


130 130

20.0 25000

20000
16.0
North

North
15000
10.0
10000

6.0
5000

0.0 0.0 0.0 0.0


0.0 East 130 0.0 East 130

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Cross Sections
J-Slice 25 J-Slice 75
40.0 40.0
Vertical

Vertical
0.0 0.0 20.0
0.0 East 130 0.0 East 130
15.0
10.0
5.0
J-Slice 60 J-Slice 100 0.0
40.0 40.0
Vertical
Vertical

0.0 0.0
0.0 East 130 0.0 East 130

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
Permeability Modeling /
Annealing
• Simulated annealing:
– honors local data
– accounts for histogram, variogram, and cross plot
– allows the integration of other types of data, e.g., seismic, welltests,
production history, ...
– solves problems that are intractable with alternative better
– understood methodologies
– easy to explain
– not multiGaussian
– requires some tradecraft in its implementation to achieve acceptable CPU
times and to avoid artifacts
– not as elegant as other methodologies
• SASIM program in GSLIB 2.0
• Covariate, continuity of extremes, non-linear averaging, ...

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
G
eo
s t
a
va tist
A
G
ila ic SL
bl al
ea So IB
Parameter File (1) tw
w
w
ftw
.G re
a
LS LI
IB B
.co rar
Simulated Annealing Based Simulation m y
************************************
START OF PARAMETERS:
1 1 1 0 0 \ components: hist,varg,ivar,corr,cpdf
1 1 1 1 1 \ weight: hist,varg,ivar,corr,cpdf
1 \ 0=no transform, 1=log transform
1 \ number of realizations
50 0.5 1.0 \ grid definition: nx,xmn,xsiz
50 0.5 1.0 \ ny,ymn,ysiz
1 0.5 1.0 \ nz,zmn,zsiz
69069 \ random number seed
4 \ debugging level
sasim.dbg \ file for debugging output
sasim.out \ file for simulation output
1 \ schedule (0=automatic,1=set below)
...
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
G
eo
s t
a
va tist
A
G
ila ic SL
bl al
ea So IB
Parameter File (2) tw
w
w
ftw
.G re
a
LS LI
IB B
.co rar
0.0 0.05 10 3 5 0.001 \ schedule: t0,redfac,ka,k,num,Omin m y
10.0 \ maximum number of perturbations
0.1 \ reporting interval
0 \ conditioning data:(0=no, 1=yes)
../data/cluster.dat \ file with data
1 2 0 3 \ columns: x,y,z,attribute
-1.0e21 1.0e21 \ trimming limits
1 \ file with histogram:(0=no, 1=yes)
../data/cluster.dat \ file with histogram
3 5 \ column for value and weight
99 \ number of quantiles for obj. func.
1 \ number of indicator variograms
2.78 \ indicator thresholds
../data/seisdat.dat \ file with gridded secondary data
1 \ column number
...
Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada
G
eo
s t
a
va tist
A
G
ila ic SL
bl al
ea So IB
Parameter File (3) tw
w
w
ftw
.G re
a
LS LI
IB B
.co rar
1 \ vertical average (0=no, 1=yes) m y
0.60 \ correlation coefficient
../data/cal.dat \ file with paired data
2 1 0 \ columns for primary, secondary, wt
-0.5 100.0 \ minimum and maximum
5 \ number of primary thresholds
5 \ number of secondary thresholds
51 \ Variograms: number of lags
1 \ standardize sill (0=no,1=yes)
1 0.1 \ nst, nugget effect
1 0.9 0.0 0.0 0.0 \ it,cc,ang1,ang2,ang3
10.0 10.0 10.0 \ a_hmax, a_hmin, a_vert
1 0.1 \ nst, nugget effect
1 0.9 0.0 0.0 0.0 \ it,cc,ang1,ang2,ang3
10.0 10.0 10.0 \ a_hmax, a_hmin, a_vert

Cent r e f or Comput at i onal Geos t at i s t i c s - Uni v er s i t y of Al ber t a - Edmont on, Al ber t a - Canada