You are on page 1of 146

 

Formation professionnalisante  
 
Reservoir Characterization  
& Modeling 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Semaine 8 
Geostatistics 
 
 
 
 
 

Une formation IFP Training pour Sonatrach / IAP 
 
Une formation IFP Training pour Sonatrach / IAP

Petroleum Geostatistics
Geostatistical Methods and Tools for Reservoir Characterization
Renaud Meunier (Geovariances)
1st ‐ 5th December 2013

Sonatrach / IAP

Regarding this presentation

 This course material is a result of the collaboration between:


• Geovariances
• IFP Training
• with contribution of Brigitte Doligez © 2013 ‐ IFP Training

Sonatrach / IAP 2
Course objectives

 Presentation of the main concepts and tools of geostatistics for


reservoir characterization, aiming at:
• Populating a reliable distribution of petrophysical parameters
between wells using all relevant information (wells and seismic)

© 2013 ‐ IFP Training
Sonatrach / IAP 3

Course Outline
 Introduction
• Key of success for a reservoir study
• Characterization main points
• Main workflow for geomodeling
• Which data / Which tools

 Fundamentals of Geostatistics

 Geostatistical Gridding

 Geostatistical Simulations
©  2013 ‐ IFP Training

 Integrated studies

Sonatrach / IAP 4
Introduction
Key for success for reservoir study

©  2013 ‐ IFP Training
Sonatrach / IAP 5

© 2013 ‐ IFP Training

Understand
Imbricate notions (Characterize)
Modeling
Pictures from RCM_WORKFLOW_presentation 6
Sonatrach / IAP
Key for success: Two steps

 Two main steps:


1. Characterization
Determine conceptual models for each topic, to determine, parameters
and path for modeling. (e.g. Structural model, Sedimentological model…).

2. Modeling

 Use parameters resulting from characterization step to build a


digital model

 Probabilistic models are used to quantify the uncertainty in the


representation of the reality

© 2013 ‐ IFP Training
Sonatrach / IAP 7

© 2013 ‐ IFP Training

Sonatrach / IAP
Introduction
Characterization
main points

©  2013 ‐ IFP Training
Sonatrach / IAP 9

Characterization step 

MAIN IDEA

 Perform data analysis to understand your reservoir before


modeling step.

 Do not model anything if you don’t have any idea about the
results !
© 2013 ‐ IFP Training

Sonatrach / IAP 10
Tools for characterization step 

 All tools for quality control and data analysis are reliable

 Geology:
• Core description
• Log analysis (correlation, sequence stratigraphy, electrofacies)
• Statistical data analysis
• Geostatistical data characterization

 Geophysics:
• Conventional seismic facies analysis
• Seismic quality synthesis using geostatistics
• Seismic facies analysis

© 2013 ‐ IFP Training
Sonatrach / IAP 11

Definitions

 Statistics
• a branch of applied mathematics concerned with the collection and
interpretation of quantitative data and the use of probability theory
to estimate population parameters

 Geostatistics
• Branch of statistics that applies to data that are distributed spatially
© 2013 ‐ IFP Training

Sonatrach / IAP 12
Results: Conceptual model

Structural model Stratigraphic model Sedimentological model Fracture model

RT1
RT2
K
RT3 
Uncertainties

© 2013 ‐ IFP Training
table
Diagenesis model Heterogeneity Fluid model
/ fluid flow
model
Pictures from RCM_WORKFLOW_presentation 13
Sonatrach / IAP

Reservoir modeling generalities

 Building the geological model

• Coherency check of the available data


• Prediction of the spatial distribution of depositional facies, using
sequence stratigraphy concepts
• Layering, based on correlations, in units supposed to behave
homogeneously relatively to fluids flow
• Gridding, depending on heterogeneity sizes
• Populating the grid with rock‐types, , K
− Kr and Pc are defined by rock‐types
© 2013 ‐ IFP Training

Sonatrach / IAP 14
Heterogeneities

Nature and spatial repartition of


Description of
heterogeneities
heterogeneities
• geology
• seismic

consistency?

Impact of heterogeneities
Caracterization of • petrophysics
heterogeneities • well tests
• production logging

© 2013 ‐ IFP Training
Sonatrach / IAP 15

Heterogeneities Causes

 Reservoir heterogeneities result from the combination of 3


factors
• Sedimentary process (deposition)
• Diagenetic evolution (thermodynamical evolution, fluid‐rock
interactions)
• Mechanical stress (tectonics)

 Important remark: homogeneity is relative to


• Observation scale
• Diagnostic tool
© 2013 ‐ IFP Training

Sonatrach / IAP 16
Fundamental heterogeneity

 It is the one, which impacts the fluid movement, and therefore


the recovery, and the recovery mechanism
• Depletion
• Water injection
• Gas injection different
fondamental
• EOR heterogeneities
 This fundamental heterogeneity is not necessarily
• The most frequent in the reservoir
• The easiest to characterize
• The same during the whole field life

© 2013 ‐ IFP Training
Sonatrach / IAP 17

Static / Dynamic heterogeneities

 Static
• The chronological reconstruction of the sedimentary process allows
to predict the spatial organization of the depositional heterogeneities
(sequence stratigraphy)

 Dynamic
• The reconstruction of the behavior of the reservoir allows to
understand and to predict the spatial organization of the
heterogeneities for fluid flow
© 2013 ‐ IFP Training

Sonatrach / IAP 18
Introduction
Main workflow for geomodeling

©  2013 ‐ IFP Training
Sonatrach / IAP 19

Workflow for geomodeling 

Structural Stratigraphic
model model 

Reservoir grid Upscaling

Well and 
seismic
data

Proportions of facies Geological model: Flow simulation


Facies, porosity, 
permeability
© 2013 ‐ IFP Training

Integration
Integration Production 
of 4D seismic
of production forecast 
data
data

Sonatrach / IAP 20
A simplified flowchart

Database Quality Control

Log Fluid Studies Seismic


Sedimentology
Interpretation (Well test, PVT) Interpretation

Data Integration

Building of the Geological Model

Deterministic Stochastic

Oil in place Computation

Upscaling

© 2013 ‐ IFP Training
Production model

Sonatrach / IAP 21

Which data?
Which tools?
© 2013 ‐ IFP Training

Sonatrach / IAP 22
Which available data? 

Initial step

W1 W2 W3
*Data are:
Porosity
Permeability
Lithofacies
Electrofacies
Petrofacies
Rocktype
Data* only at the intersection 
between model and well 
trajectory

© 2013 ‐ IFP Training
Sonatrach / IAP 23

Which tools to populate the model?

Initial step Final step

W1 W2 W3
Tools to populate 
model?

Petrophysical data
Data only at the intersection 
between model and well trajectory
© 2013 ‐ IFP Training

Sonatrach / IAP 24
Which tools to populate the model?

Initial step Final step

W1 W2 W3
Tools to populate 
model?

Petrophysical data
Data only at the intersection 
between model and well trajectory • Mapping for each layer
e.g. Interpolation
• Geostatistics tools

© 2013 ‐ IFP Training
e.g. Kriging, Simulation (SGS,SIS,Turning Band…)

Sonatrach / IAP 25

Which data to constrain the model results?

Initial step Final step

Petrophysical data
Data only at the intersection 
between model and well trajectory
Mapping tools
Porosity
Permeability Geostatistics tools
Lithofacies
Electrofacies Seismic
Petrofacies
© 2013 ‐ IFP Training

Rocktype

Can be used to constraint
Geology mapping or geostatistics
Sonatrach / IAP 26
Chapter I: Basic statistics

©  2013 ‐ IFP Training
Sonatrach / IAP 27

Basic statistics analysis

Cross plot
Base Map
© 2013 ‐ IFP Training

Sonatrach / IAP 28
Basic statistics analysis: Histograms
Count: 2200 sample (porosity data)

 Histogram  Cumulative Histogram

frequency of 
occurrence in each
interval

median (easier to read
on CDF than 
on histogram)

© 2013 ‐ IFP Training
mean: m= 1800 median = 1900

Sonatrach / IAP 29

Definition of the central value

Definition of central values


 Arithmetic mean= sum/count (sensitive to extreme or “anormal “
value)
 Median =value dividing the population in 2 halves (not sensitive
to extreme value)
• the median: value corresponding to a cumulated frequency of 50%
more generally the quantiles: values corresponding to a specified
frequency

 Mode =most frequent value (may not be unique*)


© 2013 ‐ IFP Training

 Ex: {1,1,5,6,7}: 1 2 3 4 5 6 7
Mode Median
Arithmetic mean

Sonatrach / IAP 30
Basic statistics analysis

1 N

 2= variance =
N
[ z
1
i  m]2 = E{(X‐mx)2}

high  data are
heterogeneous

 Standard deviation:  = 2 small  data are


homogeneous

 Coefficient of variation: /m: length of tail of a distribution


 Dispersion: The variance and the standard deviation are measures
of the dispersion of the outcomes relative to the mean value

© 2013 ‐ IFP Training
Sonatrach / IAP 31

Basic statistics analysis

 Different statistics of order 1

1 N
– Arithmetic mean: maz   z ( xi )
N i 1
– Geometric mean: mgz  N  iN1 z ( xi )

– Harmonic mean: N
mhz  N
1
( )
i 1 z ( xi )
© 2013 ‐ IFP Training

– NB: mhz≤mgz≤maz

Sonatrach / IAP 32
Basic statistics analysis

 Applications: permeability upscaling

Ky
 K+
Kx
arithmetic harmonic
averaging averaging
flow direction by colums

Ky
0 K-
Kx
harmonic arithmetic

© 2013 ‐ IFP Training
averaging averaging
by rows
K-  K  K+
Sonatrach / IAP 33

Basic statistics analysis

 Data distribution
• Number of defined samples
Mode
• Minimum and maximum values average
(mean)
• Mean / quantiles / mode

 Variability of data
• Variance / standard deviation
• Inter‐quartile range dispersion

 Shape of distribution
• Skewness (dissymetry)
• Kurtosis
© 2013 ‐ IFP Training

• Coefficient of variation
(Mode = most frequent value of the distribution) 

Sonatrach / IAP 34
Dispersion

 Caution 1
• Dispersion is number‐of‐classes dependant

© 2013 ‐ IFP Training
Sonatrach / IAP 35

Dispersion
 Caution 2
• Dispersion is volume window dependant (support effect)

© 2013 ‐ IFP Training

Sonatrach / IAP 36
Probability density function

Normal density Log normal density

© 2013 ‐ IFP Training
Sonatrach / IAP 37

PDF/CDF

 PDF / CDF
probability

1
Cumulative density
F(x) = P(X<x)
function (CDF)

Probability density frequency


f(x) = F’(x)
function (PDF)
© 2013 ‐ IFP Training

Sonatrach / IAP 38
Gaussian Distributions

 The distribution is entirely characterized by its mean m and its variance 2

 In the particular case m is 0 and 2 is 1 we say the distribution is normal

 It is sometimes called the bell shaped probability density

 Its properties explain the wide range of its applications

© 2013 ‐ IFP Training
Sonatrach / IAP 39

Exercise

 Draw the probability law of the sum of outcomes from tossing 2


dices
PROBABILITY LAW

© 2013 ‐ IFP Training

1/36 2/36 3/36 4/36 5/36 6/36 5/36 4/36 3/36 2/36 1/36

Sonatrach / IAP 40
Multivariate statistics

 Scatter plot
• Indispensable for a visual inspection of the joint behavior of two
variables (correlations / conditional dependence...)

f
thick
0.20
300

f 0.15
250
0.20 0.10
200
0.15 0.05
150
0.10
100 100 150 200 250 300 thick

© 2013 ‐ IFP Training
0.05
5 10 15 20 25 phi

5 10 15 20 25 phi

Sonatrach / IAP 41

Multivariate statistics

Covariance / correlation coefficient

 Covariance C(X,Y) = Cov(X,Y): gives the joint variations of X and Y


around their means
C  X ; Y   E  X  m X Y  m Y 
 x  y i  m Y 
1
 i  mX
n i

 Correlation coefficient r(X;Y) gives the linear trend of the


relationship between X and Y
C X ;Y 
© 2013 ‐ IFP Training

 X ;Y  
 X Y

Sonatrach / IAP 42
Multivariate statistics

Covariance / correlation coefficient

 Fundamuntal properties for Covariance C(X,Y): basis for kriging


computations

var  X  Y   var  X   var Y   2 C  X ; Y 

 X and Y independent C(X;Y)=0

© 2013 ‐ IFP Training
Sonatrach / IAP 43

Multivariate statistics

 Covariance / correlation coefficient


Depending on the shape of the
cloud of points and of their
thick relative location compared to
both means, the covariance will
300
- + + + be positive, negative or null
250
my=thickmoy

yi-my >0

200
yi-my <0
150 The covariance is affected by the
+ - magnitude of the data values.
- -
100
© 2013 ‐ IFP Training

If both x and y values are multiplied


by k, the covariance increases by a
5 10 15 20 25 phi
factor of k2.
xi-mx <0 xi-mx >0
mx= poromoy 44
Sonatrach / IAP
Multivariate statistics

 Covariance / correlation coefficient


• Main pitfalls: allways look at scatter plots!

+
+ +
+

© 2013 ‐ IFP Training
Correlation reduced from 0.9 to 0.4 Apparent correlation of 0.9 caused by the 3 outliers
due to 2 outliers... BUT most data are not correlated

Sonatrach / IAP 45

Multivariate statistics

Covariance / correlation coefficient


 Main pitfalls: always look at scatter plots!

 Low or null correlation coefficients do not necessarily imply


© 2013 ‐ IFP Training

independence

Sonatrach / IAP 46
Multivariate statistics

Linear regression

 Best Linear Unbiased Estimator

Y
y  ax  b
 C X ;Y 
 a 
with  X
 b  m  am
 Y X

© 2013 ‐ IFP Training
X
Sonatrach / IAP 47

Multivariate statistics

Linear regression

 Unbiased: E(y‐y*)=0

 Best: Minimal estimation error variance: E[(y‐y*)2] min

y*
y
© 2013 ‐ IFP Training

x X
Sonatrach / IAP 48
Chapter II: 
Fundamentals of Geostatistics

©  2013 ‐ IFP Training
Sonatrach / IAP 49

History

 Pionneering work in South Africa gold mines [Krige, Sichel, de


Wijs]

 60's: Fundations of mathematical theory [Matheron]

 70's: Dissemination around the world (Stanford university, South


Africa geostatistical school, Norwegian center); development of
estimation (kriging) and simulation methods

 80's: Object based methods (Norway); uncertainties

 90's: Pluri‐gaussian methods

 00's: Multi‐Point geostatistics


© 2013 ‐ IFP Training

Sonatrach / IAP 50
Limitations of Classical Statistics 

 These two maps have same histograms but different spatial


characteristics. We need therefore other tools (than global
statistical tools) to evaluate spatial variability

© 2013 ‐ IFP Training
Sonatrach / IAP 51

Basic concepts

 Random functions (RF) are functions Z(u), which are defined both
in a geographic space (physical reality) and in a probabilistic space
(mathematical model)

 Locally, the random function is a random variable

 For each pair of points u and u+h, Z(u) and Z(u+h) are not
independent, but are linked by a correlation, which expresses the
spatial structure of the RF
© 2013 ‐ IFP Training

?
? ?
? ?
?
z x
Sonatrach / IAP 52
Basic concepts

 Random functions
• Interpretation of the spatial distribution of a regionalized variable
z(u) as one realization of a random function Z(u)
• At one point u where no measurement is available, the values Z(u)
are unknown, but well defined

 2 levels of abstraction
• 1st level: Regionalized variable = mathematical function
• 2nd level: Regionalized variable = one realization of a random function

© 2013 ‐ IFP Training
Sonatrach / IAP 53

The Geostatistical Framework

 The real phenomenon is considered as a function z(x) depending


on the location x:

This function is called a regionalized variable

 The regionalized variable is interpreted as an outcome of a


random function Z(x)
© 2013 ‐ IFP Training

Sonatrach / IAP 54
Stationarity

 How to make statistics from an unique outcome of a random


process?

 A stationarity hypothesis should be made


• The variable has certain characteristics not dependent on the
position in the space

 Stationarity of order 2 hypothesis: the mean and the variance are


invariant in the space

© 2013 ‐ IFP Training
Sonatrach / IAP 55

Basic concepts

Order 2 stationarity

 The moments of order 1 and 2 of Z(u) are invariant under


translation
• E[Z(u)]=m(u)=m
• Cov(Z(u);Z(u+h))= E[Z(u)Z(u+h)]‐m2=C(h)
• Var[Z(x+h)‐Z(x)] =2 (h)
© 2013 ‐ IFP Training

Sonatrach / IAP 56
Basic concepts

By introducing an hypothesis of stationarity

 It will be possible to get from the data replicates of the statistical


features, particularly the variability between 2 points

© 2013 ‐ IFP Training
Sonatrach / IAP 57

Basic concepts

Non‐stationarity

 Mathematically: gradual trend in data

 Geologically: trend in the reservoir geology (ex: marine


transgression or regression)

 Stationarity depends on scale…

HeightHeight
= non= stationary
stationary
© 2013 ‐ IFP Training

Sonatrach / IAP 58
Chapter II: 
Fundamentals of Geostatistics
Geostatistical tools
to quantify the spatial variability

©  2013 ‐ IFP Training
Sonatrach / IAP 59

How to quantify the spatial variability?

The variogram: main tool of geostatistics

 Definition
• Basic tool to analyze the special variability
• Variogram aims at capturing the regional organization of data, which
is not purely random
• It is linked with the measure of spatial correlation between data
values separated by given distance, and reflects the intuitive idea
that data values are generally more correlated for short distances
than for long distances
• Variogram is used in estimation procedures such as kriging,
introduced by D. Krige and formalized by G. Matheron
© 2013 ‐ IFP Training

Sonatrach / IAP 60
The experimental variogram: Example

35 35 33 33 34 31 35 37 41 41

h =100 m

1 N
 (h)  *  Z ( x  h )  Z ( x ) 
2
General formulation: 2 * N ( h) 1

1
 (100 m)  ( 0 + 4 + 0 + 1 + 9 + 16 + 4 + 16 + 0) =2.77
2*9

© 2013 ‐ IFP Training
Sonatrach / IAP 61

The experimental variogram: Example

35 35 33 33 34 31 35 37 41 41

h =200 m

1 N
 (h)  *  Z ( x  h )  Z ( x ) 
2
General formulation
2 * N ( h) 1

1
 (200 m)  ( 4 + 4 + 1 + 4 + 1 + 36 + 36 + 16 )=6.375
© 2013 ‐ IFP Training

2 *8

Sonatrach / IAP 62
The experimental variogram: Example

 Unidirectional horizontal variograms with lag = 100 and 200 m.


The variogram cloud is displayed

50 50

40
40

Variogram : var#1
Variogram : var#1

30
30

20 2
20 2 1 4
3 6
4
6
5

© 2013 ‐ IFP Training
10
10 8
8

9
0
0 100 200 300 400 500 600 700 800
0
0 100 200 300 400 500 600 700 800 900
Distance (m)
Distance (m)

Sonatrach / IAP 63

Experimental variogram computation

110. 120. 130. 140.


515. 515.

510. 510.
20.
505. 505.

500. 500.
0.5*[z(x+h)-z(x)]**2

15.
495. 495.

490. 490.

10.
485. 485.

110. 120. 130. 140.

5.

Base map of n points
0.
0. 1. 2. 3. 4.
© 2013 ‐ IFP Training

Distance (Kilometer)

Variogram cloud n x (n‐1)/2 pairs 

Sonatrach / IAP 64
Experimental variogram computation

X (Kilometer) Distance (Kilometer)


110. 120. 130. 140.
515. 515. 0. 1. 2. 3. 4.
20. 20.
510. 510.

505. 505.

Y (Kilometer)
Y (Kilometer)

500. 500. 15. 15.

495. 495.

Variogram
Variogram
490. 490.
10. 10.
485. 485.

110. 120. 130. 140.


X (Kilometer)

5. 5.
Base Map

© 2013 ‐ IFP Training
0. 0.
0. 1. 2. 3. 4.
Distance (Kilometer)

Variogram Cloud
Sonatrach / IAP 65

Experimental variogram computation

Distance (Kilometer)
0. 1. 2. 3. 4.
20. 20.

2 x dl
15. 15.
Distance parameters:
Variogram

‐ Lag value:
Variogram

10. 10.
‐ Tolerance: dl
‐ Number of lags
5. 5.
© 2013 ‐ IFP Training

0. 0.
0. l
1. 2l
2. 3l
3. 4l
4.
Distance (Kilometer)

Sonatrach / IAP 66
Practical variogram computation

 Variograms on 3D data
The pairs of data are assigned to angular sectors and classes of
distances
• For a specific distance (multiple of the lag), Point A of the pair (O, A)
is included in the above volume
• For an omnidirectional 
analysis, this volume is 
a disc lag
e
nc
er a
• In 2D, the value Z” is not  lag lag tol

2x
taken into account

2 x Z”
lag

© 2013 ‐ IFP Training
O
 = Angular Tolerance (Cone Angle) 67
Sonatrach / IAP

Recommendations in Variograms calculations

 Display the number of pairs

 Look at possible anisotropies


• Calculate the variogram map when possible
• Calculate directional variograms

 Check possible outliers with the variogram cloud

 Make a few trial and errors to the parameters

 Limit the number of lags because the experimental variogram is


biased at distances larger than 1/3 of the field extension
© 2013 ‐ IFP Training

Sonatrach / IAP 68
Spatial variability

 The behaviour near the origin of the variogram is related to the


spatial continuity of the variable
(h) (h)
discontinuous
random

h h

(h) (h)
highly continuous
continuous

© 2013 ‐ IFP Training
h h

Sonatrach / IAP 69

Spatial variability

 The behaviour at long distance of the variogram is linked to


stationarity

(h) (h) (h)

no trend
presence of a trend presence of a periodicity
h h h
© 2013 ‐ IFP Training

Sonatrach / IAP 70
Spatial variability

Non stationarity

 Mathematically
• Spatial variation of the mean
• Gradual trend in the data values

 Geologically
• A trend in the reservoir geology
− Transgression / Regression
− Change in depositional environment

© 2013 ‐ IFP Training
Sonatrach / IAP 71

Spatial variability

Non stationary case: the variogram does not exist

 A spatial phenomenon can be modeled as the sum of two terms:


a regional term and a residual term

Z(x)= R(x) + m(x) + ε(x)


HF BF noise
stationary trend
deterministic
© 2013 ‐ IFP Training

Sonatrach / IAP
no trend: stationary variable trend: non stationary variable 72
Stationarity

 Stationarity can be a reasonable choice


• At some scales, not others
• For some variables, not others
• For some reservoirs, not others

 This choice strongly impacts the results of the modeling

 Stationarity is an almost‐required choice in geostatistics

 In practice: local stationarity over moving neighborhood

© 2013 ‐ IFP Training
Sonatrach / IAP 73

The Variogram Model

 The variogram model (h) characterizes the spatial correlation


of the underlying Random Function

 It is a Unique function matching all experimental variograms


calculated in different directions
© 2013 ‐ IFP Training

Sonatrach / IAP 74
Why a Variogram Model?

 Kriging requires knowledge of correlation function for all ranges


and azimuths

 Smoothes the statistical fluctuations

 Ensures positive estimation variances (only certain mathematical


functions, called authorized variogram models, satisfy this
condition)

© 2013 ‐ IFP Training
Sonatrach / IAP 75

Features of the Variogram Model

 The model smoothes the statistical fluctuations

sill ~ data variance

Range(zone of influence)
(h)

nugget effect
© 2013 ‐ IFP Training

Distance

Sonatrach / IAP 76
Link between variogram and covariance

 C(h)= E[(Z(x+h)‐m)(Z(x)‐m)]

 (h)=C(0)‐C(h)
70. 50.

60.
40.
50.

40. 30.

30.
20.

20.
10.
10.

0. 0.
0. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 0. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.
Distance (Kilometer) Distance (Kilometer)

© 2013 ‐ IFP Training
Variogram Covariance

Sonatrach / IAP 77

Nugget effect

0. 1. 2. 3. 4. 5. 6. 7. 8.

1.5 1.5
1/2 square difference of values

sill

1.0 1.0

nugget effect
0.5 0.5

Range ~6
© 2013 ‐ IFP Training

0.0 0.0
0. 1. 2. 3. 4. 5. 6. 7. 8.
Lag = distance between samples
Variogram porosity (omni) - Exp

Sonatrach / IAP 78
Comparison

Pure nugget model Gaussian model

© 2013 ‐ IFP Training
Spherical model Exponential model
Sonatrach / IAP 79

Geometric Anisotropy

Distance (Kilometer)
0. 1. 2. 3. 4. 5.

1.00
Sill Direction 2 1.00
Direction 1

0.75 0.75
Gamma

0.50 0.50

0.25 0.25

Range 1 Range 2
0.00 0. 1. 2. 3. 4. 5. 0.00
Distance (Kilometer)
© 2013 ‐ IFP Training

The ranges are located on an ellipse 
(or an ellipsoid)

Sonatrach / IAP 80
Zonal Anisotropy

Distance (Kilometer)
0. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.
Sill 2
2.0 2.0
Direction 2

1.5 1.5

Gamma
Sill 1
1.0 1.0
Direction 1

0.5 0.5

Range
0.0 0.0
0. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

© 2013 ‐ IFP Training
Distance (Kilometer)
Cross section in a 
sedimentary formation

Sonatrach / IAP 81

Geostatistical models and stationarity

 Three types of Random Functions may be modelled


• RF strictly stationary: the mean is constant and the variance exists.
The variogram is bounded
• RF intrinsically stationary: the mean of the increments is constant
and the variance of increments exists. The variogram exists but may
be not bounded
• RF non stationary: the mean is not constant and the variance does
not exist. The variogram does not exist

 The trend can be removed from the data and analysis is


performed on stationary residuals
© 2013 ‐ IFP Training

Sonatrach / IAP 82
Chapter III: 
Kriging

©  2013 ‐ IFP Training
Sonatrach / IAP 83

Deterministic methods

 Draw a map
• Geologist draw a map based on his understanding of the reservoirs
rocks distribution Drawn map

 Build a map using mathematical tools


• E.G.: Least square or average interpolation

data
© 2013 ‐ IFP Training

interpolation

Sonatrach / IAP 84
Deterministic methods

 Draw a map
• Geologist draw a map based on his understanding of the reservoirs
rocks distribution

 Build a map using mathematical tools


• Least square
Well data

Gridding with well constraint

Seismic data

© 2013 ‐ IFP Training
Sonatrach / IAP 85

Gridding methods

 Behind each model there is a decision

 All models fit the data

0.23 0.23 0.23 0.23 0.23


x x x x x 0.23
x
0.06 0.06
x x 0.06
x
0.58 0.58 0.58
x 0.01
© 2013 ‐ IFP Training

x x
0.01 0.01 x
x x

Sonatrach / IAP 86
Kriging

porosity
10000
10000
12.00
9000 11.50
11.00
8000 10.50
10.00
7000
9.50
6000 9.00

Y (ft)
5000 8.50
Y (ft)

5000 8.00
7.50
4000
7.00
3000 6.50
6.00
2000 5.50
5.00
1000 0 4.50
4.00
0
0 5000 10000
kriging porosity
N/A
0 5000 10000 X (ft)
X (ft)

Input Data: Porosity from Wells Kriging=> Porosity Map


2.5
2.5

2.0
Variogram : Porosity

2.0

Variogram : Porosity
1.5
1.5

1.0
1.0

© 2013 ‐ IFP Training
0.5
0.5

0.0
0 2500 5000 7500 10000
Distance (ft) 0.0
0 2500 5000 7500 10000
Distance (ft)
Experimental Variogram
Variogram model
Sonatrach / IAP 87

Definition of the kriging interpolator

 Linear combination of the data (weighted average):

Z 0*    Z   are the kriging weights

 Unbiased: 
E Z 0*  Z 0  0 
 Optimal: Var Z 0*  Z 0   minimum
© 2013 ‐ IFP Training

The solution is then unique

Sonatrach / IAP 88
Kriging System

 In matrix notations the kriging system is:

C  Cn  1 C

     


x =
 n
Cn Cnn  Cn0

    1

[A] x [] = [B]

© 2013 ‐ IFP Training
Kriging takes into account the distances  Kriging takes into account the 
between data two by two distances between data and the 
target point
Sonatrach / IAP 89

How Kriging Works

Z3
Z2
Zo = Z
0

C11 C12 C13 1 1 C01


Z1
C21 C22 C23 1 2 C02
x =
C31 C32 C33 1 3 C03
© 2013 ‐ IFP Training

1 1 1 0  1

Known Known
Sonatrach / IAP 90
How Kriging Works

Z2 1.0
50

Z3  Covariance Model
30

(h)
 0.5
Zo
30
30 50 .30
50 
.15
0.0
Z1 0 30 50 100
distance

C 11 C
1.0 12 C
.15 13 1
.15 1 C
.30
01
C.15
21 C 22 C
1.0 23 1
.15 2 C.30
02
x =
C.15 C C

© 2013 ‐ IFP Training
31 .15 33 1
32 1.0 3 C.30
03
1 1 1 0  1
Sonatrach / IAP 91

Confidence Interval

Depth

© 2013 ‐ IFP Training

95% of probabilities the real Z lies between [

Sonatrach / IAP 92
Influence of the variogram model

Kriging Using a Spherical 
Model

Map of Porosity
Kriging Error

Model

© 2013 ‐ IFP Training
Sonatrach / IAP 93

Influence of the variogram model

Kriging Using a Spherical 
Model With a Nugget

Map of Porosity
Kriging Error

Model
© 2013 ‐ IFP Training

Sonatrach / IAP 94
Smoothing effect of kriging

10000 2.5

9000

8000 2.0

7000 Data
1.5
6000 Kriging
Y (ft)

5000
1.0
4000

3000 0.5

2000

1000 0.0
0 2500 5000 7500 10000

0 Distance (ft)
0 1000 2000 3000 4000 5000 6000 7000 8000 90001000
Porosity
X (ft)

 The variogram computed on the kriging grid has a sill that is much
lower than the variogram computed on the real data. (2 times
lower)

© 2013 ‐ IFP Training
 The range is higher on the variogram computed on the kriging
estimation

Sonatrach / IAP 95

Neighborhood

 When all data are used for kriging we perform a kriging in Unique
neighborhood

 The data used for kriging can be limited to a subset of the whole
set of data. We then perform a kriging in Moving neighborhood
© 2013 ‐ IFP Training

Sonatrach / IAP 96
Moving Neighborhood

 First the data beyond a limit radius from the target are discarded
from the kriging system. The limit distance is defined by the
envelope of an ellipse (ellipsoid)

 Within the ellipse only a subset of the data is retained. Using a


criterion of proximity to the target the selected data are the
closest to the target. The parameters for selecting the data are
• Number of data per regular sector
• Declustering radius
• Maximum number of empty sector

© 2013 ‐ IFP Training
Sonatrach / IAP 97

Moving Neighborhood

Search ellipse parameters


depend upon:
• Data density
• Variogram parameters
Spatial scale
Anisotropy ratio
Azimuth

Kriging: a Linear Estimator


Zo = l1Z1 + l2 Z2 +  lnZn

Data Points: Z1, Z2, …, Zn

Weights: l1, l2, … , ln
© 2013 ‐ IFP Training

Sonatrach / IAP 98
Indicator kriging

 Kriging can be used to "interpolate" lithotypes instead of real


variables

 Methodology

• Use an indicator transform to transform lithotype data l to numerical


data Il

• Use kriging to estimate the conditional probabilities I*l at the point to


be simulated

© 2013 ‐ IFP Training
Sonatrach / IAP 99

Indicator kriging

 Data: x  A,..., x   A,... coded by 1A ( x )  0 or 1

 Conditional probability:
P ( x0  A | x  A,..., x   A,...)  E (1 A ( x0 ) | x  A,..., x   A,...)

which is often difficult to calculate

 Indicator Kriging:
n
 n

[1 A ( x0 )]    1A ( x )  1     p A
K

 1   1 
© 2013 ‐ IFP Training

which is considered as a probability (conditional probability) but 
does not necessarily lie within [0, 1]

Sonatrach / IAP 100
Chapter IV: 
Multivariate geostatistics

©  2013 ‐ IFP Training
Sonatrach / IAP 101

Example

 Temperature is impacted by latitude and altitude

© 2013 ‐ IFP Training

Sonatrach / IAP 102
Multivariate geostatistics

 In Geostatistics, when the main variable shows a good correlation


with an auxiliary variable, it can be interesting to use this
auxiliary variable in order to improve the estimation of the main
variable

 For instance, seismic data are often used as auxiliary variables in


time to depth conversion, property mapping

 Indeed seismic data densely sampled improve the estimation of a


variable known only at well location

© 2013 ‐ IFP Training
Sonatrach / IAP 103

Multivariate geostatistics

Well data
Variogram
model

Cokriged map

Cross 
variogram Cokriging
model

Seismic data

Variogram
model
© 2013 ‐ IFP Training

CoKriging using Well and Seismic Data


Sonatrach / IAP 104
Multivariate geostatistics

 Map of the Norm AI used as an auxiliary variable correlated to


Porosity

Image
10000
0.25
9000 0.17
0.09
8000 0.02
-0.06
7000
-0.14
6000 -0.22
Y (ft)

-0.30
5000 -0.38
-0.45
4000 -0.53
-0.61
3000
-0.69
2000 -0.77
-0.84
1000 -0.92
-1.00
0
0 1000 2000 3000 4000 5000 6000 7000 8000 90001000
Norm AI N/A

© 2013 ‐ IFP Training
X (ft)

Sonatrach / IAP 105

Multivariate geostatistics

 The cokriging porosity estimation shows more variability, coming


from the seismic variable

porosity porosity
10000 10000
12.00 12.00
11.50 11.50
11.00 11.00
10.50 10.50
10.00 10.00
9.50 9.50
9.00
Y (ft)

9.00
Y (ft)

5000 8.50 5000 8.50


8.00
8.00
7.50
7.50
7.00
7.00
6.50
6.50
6.00
5.50 6.00
5.00 5.50
0 4.50 5.00
4.00 0 4.50
0 5000 10000 4.00
cokriging porosity 0 5000 10000
N/A kriging porosity
X (ft) N/A
X (ft)
© 2013 ‐ IFP Training

Sonatrach / IAP 106
Multivariate geostatistics

Data measurements are available for several correlated variables

 Two cases
• Isotopy: all variables are measured at all data points
• Heterotopy: some variables are not measured at some data points

 Objective
• Estimate (co‐kriging) a variable Zi0 at a point x0 by a linear
combination of data for a set of variables Zi, i=1, …, N

 The data values for a variable different of the estimated variable


has an indirect influence on that variable and may improve the
estimation

© 2013 ‐ IFP Training
Sonatrach / IAP 107

Multivariate geostatistics

 Structural Analysis
• Simple Variograms:

Z1(h)=1/2 E[Z1(x+h)—Z1(x)]²
Z2(h)=1/2 E[Z2(x+h)—Z2(x)]²

• Cross Variograms:

Z1 Z2 (h)=1/2 E{[Z1(x+h)—Z1(x)]  [Z2(x+h)—Z2(x)]}

This quantity can be <0. It reflects the correlation at a given distance.

• Cross Covariance:
© 2013 ‐ IFP Training

Cov Z1 Z2 (h)= E{[Z1(x)—mZ1(x)] [Z2(x+h)—mZ2(x)]}

Sonatrach / IAP 108
Cokriging of multivariate data

 The spatial cross correlation is modeled by means of the cross


variogram

 Cokriging weights are assigned to the variable of interest as well


as to the auxiliary variables

© 2013 ‐ IFP Training
Sonatrach / IAP 109

Modeling in the Multivariate Case

 We must find an authorized model, consistent for all variables

 We use a constraining but workable hypothesis:

The linear model of co‐regionalization
all simple variograms and cross variograms must be 
linear combinations of the same basic structures gu(h)

(basic structure = variogram with given range and sill of 1 .)
© 2013 ‐ IFP Training

Sonatrach / IAP 110
Simple and Cross Variograms

© 2013 ‐ IFP Training
Sonatrach / IAP 111

Simple and Cross Variograms

 Bivariate variogram: Example (Phi/Ip)


Distance (km) Distance (km)
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
0.125 0.125

0.100 0.100

0.100 0.100
Variogram : pareia
Variogram : pareia

Variogram : pareia

Variogram : pareia

0.075 0.075
0.075 0.075

0.050 0.050
0.050 0.050

0.025 0.025 0.025 0.025

0.000 0.000 0.000 0.000


0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
Distance (km) Distance (km)
Distance (km) Distance (km) Distance (km) Distance (km)
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
10 10
2500 2500 2500 2500
Variogram : pareia & imp-avg
Variogram : pareia & imp-avg

Variogram : pareia & imp-avg

Variogram : pareia & imp-avg

10 10
2000 2000 2000 2000
Variogram : imp-avg
Variogram : imp-avg

Variogram : imp-avg

Variogram : imp-avg

5 5

1500 1500 1500 1500

0 0 0 0
1000 1000 1000 1000

500 500 -5 -5 500 500


-10 -10

0 0 0 0
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
Distance (km) Distance (km) Distance (km) Distance (km)

mod1  1  cubic(1500) 2  lin(2000) 3  pep mod2  1  sph(2000) 2  sph(700)


 imp  1 1   2 2   3  imp   '1  1   '2  2   '3
© 2013 ‐ IFP Training

 phi  1 1   2 2   3  phi   '1  1   '2  2   '3


 imp / phi  1 1   2 2   3  imp / phi   '1  1   '2  2   '3

Sonatrach / IAP 112
Ordinary Cokriging

 Principle: Estimate Z1 using a linear combination of Z1 and Z2


measured at scattered data points

 The ordinary cokriging system is:

© 2013 ‐ IFP Training
Sonatrach / IAP 113

Ordinary Cokriging

Matrix notation

© 2013 ‐ IFP Training

Sonatrach / IAP 114
Self Krigeability

 A variable is self‐krigeable when its cokriging (in the


isotopic case) is identical to its kriging
• No correlation:

• Perfect correlation

• Intrinsic correlation:simple and cross variograms are all


proportional

© 2013 ‐ IFP Training
This is always the case if there is only one basic structure

Sonatrach / IAP 115

Data Integration

 Optimal use of multivariate data sets

 Mixing wells and seismic data

 Time to depth conversion


© 2013 ‐ IFP Training

Sonatrach / IAP 116
Collocated CoKriging

 When dealing with


• The main variable sampled at few wells
• A secondary variable measured on a seismic grid

Wells

Seismic grid

© 2013 ‐ IFP Training
Target node

Sonatrach / IAP 117

Collocated CoKriging

 It is very close to cokriging except that one supplementary weight


is assigned to seismic data at the target grid node

Zo = ZTT

 Note: this way of achieving collocated cokriging is known as


« extended collocated cokriging » (instead of « strictly
collocated » where only the auxiliary variable at target grid node
is used)
© 2013 ‐ IFP Training

Sonatrach / IAP 118
Collocated Cokriging

Zo = Z  T  T


Z1, T1 Z3,T3




T0 



© 2013 ‐ IFP Training
Z2,T2

Sonatrach / IAP 119

Advantages of Collocated CoKriging

 Incorporates much more information

 Carries all the detail of the spatial correlation © 2013 ‐ IFP Training

Kriging Collocated
Sonatrach / IAP 120
When to use Collocated CoKriging

There should be a significant correlation between hard and soft data

© 2013 ‐ IFP Training
Sonatrach / IAP 121

Limitations of Cokriging

 Traditional cokriging requires more modeling effort than kriging

 Still produces a smooth image of reality, but not as smooth as


kriging because of the incorporation of secondary information
© 2013 ‐ IFP Training

Sonatrach / IAP 122
Kriging vs. Collocated Cokriging

Kriging Porosity

© 2013 ‐ IFP Training
Collocated Cokriging
Porosity & AI
r = ‐0.83

Sonatrach / IAP 123

Limitations of Cokriging

 It is often difficult to fit a bivariate model on experimental


variograms and cross‐variograms

 An alternative is to model the variogram on seismic (dense data)


and make an assumption on the similar behavior of the other
variogram and cross‐variogram

Markov Bayes Assumption
© 2013 ‐ IFP Training

Sonatrach / IAP 124
Markov Bayes Assumption

 The bivariate variogram model is defined by means of these


relationships

CZ,T(h)=CZ,T(0) /CT(0)*CT(h)

CZ(h)=CZ(0) /CT(0)*CT(h)

 Requires a linear regression between well and seismic data

© 2013 ‐ IFP Training
 Not often easy to validate or choose the parameters

Sonatrach / IAP 125

Kriging with External Drift

 Basic Uses
• Time‐to‐depth mapping
• Simple data integration
− Seismic trend
− Other trend

 Requirements
• Drift Model
• Generalized Covariance Model
− A type of variogram model based on polynomials

 Automatic
© 2013 ‐ IFP Training

Sonatrach / IAP 126
Kriging with External Drift

 Non stationary technique

 The seismic provides a “shape” function with stationary residuals


on top of it

 Requirements
• Linear regression between wells and seismic
• High correlation is very important

© 2013 ‐ IFP Training
Sonatrach / IAP 127

External Drift

Seismic profile S

Trend=a+b*S

Trend+residuals

Stationary residuals
© 2013 ‐ IFP Training

Sonatrach / IAP 128
Kriging with External Drift

 It is similar to Collocated Cokriging except that the slope of


regression disappears

Z  a  b  S  Residual
with:

Z ( x0 )   i  Z i
*
K
iS1

 1 i
iS1

© 2013 ‐ IFP Training
 S S i i 0
iS1
Sonatrach / IAP 129

External Drift vs. Collocated Cokriging

 Both methods give similar results when the correlation is very


high (r > 0.85)

 Collocated cokriging respects better the details of the seismic


variability
© 2013 ‐ IFP Training

Sonatrach / IAP 130
Chapter V: 
Geostatistical Simulations

©  2013 ‐ IFP Training
Sonatrach / IAP 131

Why simulations?

 To represent the complexity of a reservoir

 To assess the impact of uncertainty of a reservoir model by


producing optimistic and pessimistic scenarios

 To honor the reservoir heterogeneities


• Reservoir architecture described by the sequence of lithological
facies
• Rock property variability within given facies

 To support non linear calculations (e.g. volumetrics)


© 2013 ‐ IFP Training

Sonatrach / IAP 132
Why simulations and not interpolation?

 An interpolation, even a geostatistical method like kriging,


exaggerates the continuity

 By adding information, the model becomes more and more


complex and realistic

 While the interpolation looks for an intermediate value,


simulations aim at reproducing the variability

© 2013 ‐ IFP Training
Sonatrach / IAP 133

Conditional Simulation

Interpretation of lithofacies from well logs (courtesy IFP/CG)

© 2013 ‐ IFP Training

Sonatrach / IAP 134
Conditional Simulation

Interpretation of lithofacies from well logs (courtesy IFP/CG)

© 2013 ‐ IFP Training
Sonatrach / IAP 135

Conditional Simulation

Interpretation of lithofacies from well logs (courtesy IFP/CG)

© 2013 ‐ IFP Training

Sonatrach / IAP 136
Conditional Simulation

Interpretation of lithofacies from well logs (courtesy IFP/CG)

© 2013 ‐ IFP Training
Sonatrach / IAP 137

Why simulations?

 Kriging does not give a realistic image of the reality

© 2013 ‐ IFP Training

Kriged porosity

Real porosity
Sonatrach / IAP 138
Simulations for reservoir characterization 

 Ignoring the reservoir heterogeneities leads to biased predictions


and poor development plans

 Example of a five spots scheme: injection at the centre, recovery


at the corners

© 2013 ‐ IFP Training
Sonatrach / IAP 139

Simulations for reservoir characterization 

 Oil recovery predictions vs reservoir heterogeneities

© 2013 ‐ IFP Training

Five spots simulation

Sonatrach / IAP 140
Simulations for volumetrics

 We could compute the volume of the reservoir above the cutoff


by applying it to:
• the kriging curve (Z*)
• Z*+  (optimistic)
• Z*‐  (pessimistic)

© 2013 ‐ IFP Training
 The 3 estimations are biased  needs of simulation

Sonatrach / IAP 141

Simulations for volumetrics

 With many simulations optimistic and pessimistic scenarios can


be quantified with good confidence

4 simulations Simulation vs. Kriging


© 2013 ‐ IFP Training

Sonatrach / IAP 142
Simulations of an island

 A bathymetric survey around the Yeu island (20 Km from the


French Atlantic Coast) has been used to estimate its surface

Courtesy of Centre de Géostatistique

© 2013 ‐ IFP Training
Sonatrach / IAP 143

Simulations of an island

© 2013 ‐ IFP Training

Sonatrach / IAP 144
Simulations of an island

 The bathymetric data have been kriged to interpolate the


elevation. Applying the sea level produces a smoothed island
outline

© 2013 ‐ IFP Training
Sonatrach / IAP 145

Simulations of an island

 9 simulated « islands » from 50

© 2013 ‐ IFP Training

Sonatrach / IAP 146
Simulations of an island

 Profiles of 9 simulated « islands » from 50

© 2013 ‐ IFP Training
Sonatrach / IAP 147

Simulations of an island

 Probability for a point to belong to the island calculated from the


50 simulations

© 2013 ‐ IFP Training

Sonatrach / IAP 148
Simulations of an island

Statistics from 50 simulations

 Real surface ……………… 23.32 Km2

 Kriged surface ……………. 22.88 Km2

 Simulated surface
• Average …………………… 23.17 Km2
• Minimum …………………. 15.24 Km2
• Maximum ………………… 31.90 Km2

© 2013 ‐ IFP Training
Histogram of surfaces (km2)

Sonatrach / IAP 149

Simulations of an island

 Statistics calculated on the 50 simulations for the area of the


island, the volume and the height above sea level

© 2013 ‐ IFP Training

Sonatrach / IAP 150
Objectives of simulations

 To construct a gridded model of the reservoir with its


petrophysical properties

 To represent correctly the geological heterogeneities (lithofacies),


then populate the lithofacies by the properties

 To quantify the uncertainty on the reserves and their future


recovery

© 2013 ‐ IFP Training
Sonatrach / IAP 151

Simulations output

 On simulations we are authorized to make statistics directly


related to the uncertainty

• Probability maps

• Quantile maps

• Risk analysis
© 2013 ‐ IFP Training

Sonatrach / IAP 152
Why simulation: Probability maps

© 2013 ‐ IFP Training
Sonatrach / IAP 153

Why simulation: Quantile maps

© 2013 ‐ IFP Training

Sonatrach / IAP 154
Simulations for risk analysis

CDF

1
0.6
probability of porosity > 10%

0 10 20 porosity
100
90
80
70
60
50
40
30
probability maps 20
10
threshold maps 0

 Risk analysis

© 2013 ‐ IFP Training
n equiprobable maps of the same property 
(same distribution, same variogram, same 
conditioning data)

Sonatrach / IAP 155

Estimation vs. simulations

Example
© 2013 ‐ IFP Training

Sonatrach / IAP 156
Estimation vs. simulations

conditional simulation kriging

output equiprobable realizations one deterministic model + s

honors wells honors wells
properties
honors histograms, variograms minimize average error

image noisy
smooth
same variability everywhere
flow simulations mapping
use
uncertainty computations

© 2013 ‐ IFP Training
volumetrics

Sonatrach / IAP 157

Estimation vs. simulations

 The geostatistical simulations consist of two operations


• A simulation step itself constrained by the variogram and trend
model
• A conditionning step to honor the data (based on kriging)

 The mean of several realizations of geostatistical simulations


tends toward the kriging estimation

 The variance of several realizations of geostatistical simulations


tends toward the kriging estimation variance
© 2013 ‐ IFP Training

Sonatrach / IAP 158
Simulation methods

2 Classes

 Continuous variable
• Ø, K, saturation…
• Depth

 Categorical variables
• Lithofacies
• Channels

© 2013 ‐ IFP Training
Sonatrach / IAP 159

© 2013 ‐ IFP Training

Sonatrach / IAP
Chapter V: 
Geostatistical Simulations
Simulations of continuous variables

©  2013 ‐ IFP Training
Sonatrach / IAP 161

Various algorithms of simulations

 Various algorithms to simulate continuous variables. The most


used ones are

• Sequential Gaussian Simulation (SGS)

• Turning Bands (TB)

• Gaussian Random Function Simulation (GRFS)


© 2013 ‐ IFP Training

Sonatrach / IAP 162
Sequential Gaussian Simulation 

 A grid node value is obtained as the sum of the kriging and a


random error:

yG  z k*   k*  N (0,1)
kriging
Random variable

 For the second grid node, we incorporate the first node to the
data set and so on

© 2013 ‐ IFP Training
Sonatrach / IAP 163

Sequential Gaussian Simulations (SGS)

 Simulation algorithm
1. Selection of a grid node randomly
17.1
19.9
30.1
26.2
16.4
Initial data (wells)

13.4
x
Already simulated nodes
24.5
21.9
11.2
Selected location
© 2013 ‐ IFP Training

Sonatrach / IAP 164
Sequential Gaussian Simulations (SGS)

 Simulation algorithm
2. Kriging conditioned by data + simulated values in the neighborhood

17.1
19.9
30.1
26.2
16.4 1.7

13.4
x
18.1

24.5
21.9

© 2013 ‐ IFP Training
11.2 Local conditional probability distribution:
Selected location Gaussian PDF with:
- mean (18.1) = kriging estimation
- variance (3.0) = kriging error variance
Sonatrach / IAP 165

Sequential Gaussian Simulations (SGS)
 Note about the neighborhood:
• The data search may be achieved like with kriging (search ellipsoid …)
with an additional parameters defining the optimum number of
previously simulated grid nodes
• A specific neighborhood is often used because it is speeding up the
calculations
Conditioning data 
migrated to grid nodes

Formerly simulated 
grid nodes

Next grid node to be 
simulated (target)

Neighborhood on data 
and already simulated 
© 2013 ‐ IFP Training

grid nodes 

Sonatrach / IAP 166
Sequential Gaussian Simulations (SGS)

 Simulation algorithm
3. Random drawing on the local conditional probability distribution

Simulated Gaussian value = 16

1.7

© 2013 ‐ IFP Training
18.1

Sonatrach / IAP 167

Sequential Gaussian Simulations (SGS)

 Simulation algorithm
4. Repeat steps 1 to 3, after incorporating the newly simulated data to
the conditioning data, until all the grid nodes have been simulated

 To obtain other equiprobable realizations of the random function


• Run the entire simulation process with a different path or with a
different sampling technique
© 2013 ‐ IFP Training

Sonatrach / IAP 168
Sequential Gaussian Simulation

Limitation of the method:

 Time consuming function of the neighborhood extension

 Based on simple kriging (SK), so strict stationarity is required

© 2013 ‐ IFP Training
Sonatrach / IAP 169

Turning Bands Method 

Principle

 To simulate first a random function on N lines independently with


a given covariance

 To get the simulated values at the target grid nodes by


considering the projection on N lines uniformly spread out in the
space
© 2013 ‐ IFP Training

Sonatrach / IAP 170
Turning Bands

 Consequence
• The three dimensional random function Y3(x) is obtained from the
simulated values Y1

1
Y3 ( M )  Y 1
Li
(M i )
N
• Hence before conditioning Y3 has a multi‐gaussian distribution

• To condition the simulation to the data values by adding to the


simulated values previously obtained the kriged error

© 2013 ‐ IFP Training
Sonatrach / IAP 171

Principle of the Conditioning 

 The conditional simulation is obtained by adding to the non‐


conditional simulated values the kriging of the difference
between the actual and simulated data:

Ysc(x)= Ys(x) + [Y(x) ‐ Ys(x)]K

 Y(x) and Ys(x) have the same variogram and the data
configuration is the same. Consequently the kriging weights are
the same when applied to the real or to simulated values
© 2013 ‐ IFP Training

Sonatrach / IAP 172
Conditional Simulations 

Non Conditional Simulation 

= Conditional Simulation 

+
Kriging of Residuals

© 2013 ‐ IFP Training
Sonatrach / IAP 173

Gaussian Anamorphosis

 Works only with Gaussian variables


• Perform a Gaussian anamorphosis

© 2013 ‐ IFP Training

• The n data are ordered by increasing values
Sonatrach / IAP 174
Gaussian Anamorphosis

 A possible model is the decomposition of  into orthogonal


Hermite polynomials:

 In practice the expansion is stopped at a degree N:

© 2013 ‐ IFP Training
Sonatrach / IAP 175

Gaussian Anamorphosis 

 Example of histogram modeling using the expansion of the


Gaussian anamorphosis into 30 Hermite polynomials

© 2013 ‐ IFP Training

Histogram of porosity

Sonatrach / IAP 176
Non Stationary Simulations  

 The Turning bands method apply for non stationary variables


(with trend)

 SGS does not because it requires strict stationarity


• Problem: the histogram can not be used to derive the distribution of
a non stationary variable, from a theoretical point of view
• In practice, when the histogram is clearly not Gaussian, it is still
better to anamorphose the data prior to simulate

© 2013 ‐ IFP Training
Sonatrach / IAP 177

Mixing Wells and Seismic Data

Idea

 To use the dense seismic information in addition of the sparse


wells data

Two methods

 Simulation with seismic as external drift


• The seismic attribute is considered as explaining the trend of the
wells property

 Collocated co‐simulations
© 2013 ‐ IFP Training

• The seismic attribute and the wells property are spatially correlated

Sonatrach / IAP 178
Simulation with External Drift

 Simulation of a non stationary random function

 Conditioning using kriging with external drift: the kriging weights


are calculated in order to filter out the drift

© 2013 ‐ IFP Training
Sonatrach / IAP 179

Collocated Co‐simulations

 Co‐simulation of the wells property using a linear model of


coregionalization with the seismic

 The conditioning is made by collocated cokriging


© 2013 ‐ IFP Training

Sonatrach / IAP 180
Chapter V: 
Geostatistical Simulations
Simulation of lithofacies

©  2013 ‐ IFP Training
Sonatrach / IAP 181

Simulation methods

2 Classes:

 Continuous variable
• Ø, K, saturation…
• Depth

 Categorical variables
• Lithofacies
• Channels
© 2013 ‐ IFP Training

Sonatrach / IAP 182
Categorical Simulations

 Overview of different types of algorithms
• Pixel‐based
− Sequential Indicator Simulations
− Truncated Gaussian and Plurigaussian
− Conditioned by: Well data + variograms + proportions
• Object‐based
− Boolean simulations with various marks
− Sinusoids, ellipsoids, rectangles…
− Conditioned by: Well data (+‐) + proportions
• Pattern‐based
− Multiple‐Point Statistics
− Conditioned by: Well data + Training image + 
proportions
• Process‐based

© 2013 ‐ IFP Training
− Flumy: meandering channels
− Conditioned by: Well data (+‐) + physical processes + 
proportions
Sonatrach / IAP 183

Object oriented methods

 Marked points process


• Randomly generate marks
− E.g. Poisson point process
• Random token model
• Boolean model
• Dead leaves

 Birth and death process


• Points appear or disappear with given probabilities
© 2013 ‐ IFP Training

Sonatrach / IAP 184
Boolean model

 Application to simulate channels, crevasse splays, river meanders,


bars…

© 2013 ‐ IFP Training
Sonatrach / IAP 185

Boolean model

 Application to simulate channels, crevasse splays, river meanders,


bars …

© 2013 ‐ IFP Training

Sonatrach / IAP 186
Boolean non stationary

Making vary the density of the Poisson points

Intensity=0.01

Intensity=0.02

© 2013 ‐ IFP Training
Intensity=0.05

Sonatrach / IAP 187

Pixel oriented methods

 Multiple points simulations

 Methods based on indicators


• Sequential Indicator Simulations
• Truncated Gaussian
• Truncated Pluri‐gaussian
© 2013 ‐ IFP Training

Sonatrach / IAP 188
Multiple‐Points Simulations

 Reconcile the flexible data conditioning achieved by pixel‐based


methods with the realistic shape information captured by
Boolean methods

 Learn multiple‐point statistics from a geological training image.


Then a pixel‐based simulation is applied to create facies
simulations

 Facies simulations are conditioned to well data and statisticaly


reproduce the characteristics inferred from the training image

© 2013 ‐ IFP Training
Sonatrach / IAP 189

Multiple‐Points Simulations

 A random path is used to review all the grid nodes which have not
been simulated yet

 When the target node is selected, a search phase is carried out in


order to search for the neighboring nodes already simulated or
for data. The neighboring nodes together with the target node
constitute the searched pattern

 The training image is scanned in order to find the patterns which


match the searched pattern (same facies as those of the searched
pattern): they are called the matching patterns
© 2013 ‐ IFP Training

Sonatrach / IAP 190
MPS

Matching Searched 
Patterns Pattern

 Patterns 1, 2, 3: OK

 Pattern 4: Discarded

© 2013 ‐ IFP Training
Sonatrach / IAP 191

MPS

Draw an uniform value 


between 0 and 1

If drawn value > 2/3: Orange 
If drawn value < 2/3: Green
© 2013 ‐ IFP Training

Sonatrach / IAP 192
Sequential Indicator Simulation

 SIS Algorithm

• The points are scanned according to a random path

• For each point an indicator kriging is performed using the indicators


of the data AND of the previously simulated points. It results a
cumulative probability function for the occurrence of the set of facies

• A random number is drawn uniformly between 0 and 1. By


comparing that number with the cumulative probability, the
simulated facies is assigned to the rank of the interval to which the
random number belongs

© 2013 ‐ IFP Training
Sonatrach / IAP 193

Sequential Indicator Simulation

Cdf

1
Facies 2

u  0,1
IK(x)
Facies 1
© 2013 ‐ IFP Training

Sonatrach / IAP 194
Sequential indicator Simulations (SIS)

treat x as hard data



and select a new random
  actual well data location
x : shale
  x : sand
U < p*  shale

U  p*  sand

- Estimate the conditional probability that the


point is shale: p*
- Classify point by comparing estimate to a
uniform random number from [0,1]: U  [0,1]

© 2013 ‐ IFP Training
Sonatrach / IAP 195

Sequential indicator Simulations (SIS)

 SIS is a very flexible method. However its use requires some


approximations (conditioning from neighboring data only, local
corrections)…
• For example, the only theoretical problem with this method is that I
can be less than 0 or greater than 1. When this occurs, it will be
obviously set to 0 or 1 accordingly, but the covariance is then no
longer reproduced exactly
© 2013 ‐ IFP Training

Sonatrach / IAP 196
Sequential indicator Simulations (SIS)

 SIS characteristics
• Does not correspond to any model in particular
• Takes into account indicators (facies) proportions
• It implicitly assumes no correlation between indicators
• No constraint on the indicator variograms
• It is extremely difficult to choose a mathematically consistent model
for the indicator variograms and to define how accurate the Indicator
Kriging approximation of the conditional probability is.

© 2013 ‐ IFP Training
Sonatrach / IAP 197

Truncated Gaussian

Objective
 Simulate a discrete variable (lithofacies) at all the nodes of a 3D
grid
 Spatial distribution of simulated lithofacies must be different in
the horizontal plane and in the vertical direction (in the
chronostratigraphic space)
• Vertically: it reproduces the sedimentary process
• Horizontally: it characterize the homogeneity of the field

Basic idea
© 2013 ‐ IFP Training

 Each lithofacies indicator is considered as being the


transformation of some truncated gaussian random function

Sonatrach / IAP 198
Truncated Gaussian

 Algorithm
1. Simulate a stationary Gaussian random function Z(x) with a
covariance function C(h)

Z(x)

© 2013 ‐ IFP Training
(Matheron et al. 1987, Galli et al 1994..)

Sonatrach / IAP 199

Truncated Gaussian

 Algorithm
2. Truncate Z(x) at a desired level and assign lithofacies according to
the chosen threshold

Z(x)

threshold

x
Simulated lithofacies
© 2013 ‐ IFP Training

(proportions pk / covariance Kk(h)


Z<cutoff Z>cutoff

Sonatrach / IAP 200
Truncated Gaussian

How to choose the threshold?

Proportions for 
shale (green)

80% Pshale  G (1)  80 %

© 2013 ‐ IFP Training
10%

CDF from Y(x) Pshale  G(  1)  10 %


Sonatrach / IAP 201

Truncated Gaussian Simulation

 The thresholds are determined from the proportions of the facies

 Y(x) designates a stationary Gaussian RF, 0 mean‐variance 1

Fi x| ti 1  Y ( x )  ti 

1 if t i 1  Y ( x )  t i
1Fi ( x )=
© 2013 ‐ IFP Training

0 elsewhere

Sonatrach / IAP 202
Truncated Gaussian Simulation

 For the first facies:

p1  P ( x  F1 )  P (   Y ( x )  t1 )  G (t1 )
Where G is the cumulative Gaussian function

 For the next facies:


pi  P ( x  Fi )  G (ti )  G (ti 1 )

© 2013 ‐ IFP Training
Sonatrach / IAP 203

Truncated Gaussian Simulation

 The crucial point is the proportion of the different facies.

The proportion is constant whatever the level: 
45%‐25%‐30%
© 2013 ‐ IFP Training

Sonatrach / IAP 204
Truncated Gaussian Simulation

 When the proportions vary with the level we speak of vertical


non stationarity

pi varies with depth, so ti varies with 
depth.

© 2013 ‐ IFP Training
Sonatrach / IAP 205

Variographic analysis

 We need to have a 3D variogram model for simulating the


Gaussian Random Function

 The difficulty is that we can only calculate experimental


variograms on the indicators of the facies
© 2013 ‐ IFP Training

Sonatrach / IAP 206
Truncated Gaussian

 Effect of the variogram model C(h)


1. Simulation of the Gaussian variable

exponential model exponential model


ranges (1500;1500;10)

ranges (300;300;3)
gaussian model gaussian model

© 2013 ‐ IFP Training
Sonatrach / IAP 207

Truncated Gaussian

 Effect of the variogram model C(h)


1. Simulation of the Gaussian variable

exponential model exponential model


ranges (1500;1500;10)

ranges (300;300;3)

gaussian model gaussian model


© 2013 ‐ IFP Training

Sonatrach / IAP 208
Truncated Gaussian

 Effect of the variogram model C(h)


2. Truncation using proportions

exponential model exponential model


ranges (1500;1500;10)

ranges (300;300;3)
gaussian model gaussian model

© 2013 ‐ IFP Training
Sonatrach / IAP 209

Truncated Gaussian

Effect of the Thresholds

gaussian model , gaussian model ,


=300m =300m
proportions 2
proportions 1

gaussian model,
gaussian model gaussian model,
=1500m =1500m
© 2013 ‐ IFP Training

Sonatrach / IAP 210
Truncated Pluri‐Gaussian

Principle

 Generalization of the Truncated Gaussian model


• The thresholds are defined according to two different Gaussian
random variables (correlated or not)

 Basic idea
• The 2 different random functions may have different variogram
(ranges, anisotropy)
• It allows to mimic complex facies distributions resulting from multi‐
stage processes (sedimentation / diagenesis / erosion / intrusion...)

© 2013 ‐ IFP Training
Sonatrach / IAP 211

Truncated Plurigaussian Functions

 In practice 2 GRF are enough to give enough flexibility,


particularly in reproducing transitions between the different
facies

 The following scheme is also called Lithotype rules. It represents


the rule for assigning the lithotypes when applying thresholds on
Gaussian Function 1 (along x axis) and on Gaussian Function 2
(along y axis)
© 2013 ‐ IFP Training

Sonatrach / IAP 212
Pluri‐Gaussian Simulation (PGS)

Threshold to translate lithotype distribution integrating on


geological constraint

Max.G1

Threshold
T1
Min. G1

First Gaussian variable 
(G1) Threshold Resulting 
T2 lithofacies 
Min. G2 Max. G2 simulation

© 2013 ‐ IFP Training
Second Gaussian variable (G2)
Sonatrach / IAP 213

Pluri‐Gaussian Simulation (PGS)

Different variogram types give different body contours

Max.

Threshold

Min.
© 2013 ‐ IFP Training

Sonatrach / IAP 214
Pluri‐Gaussian Simulation (PGS)

Gaussian Exponential Spherical

© 2013 ‐ IFP Training
Lithofacies simulations

Sonatrach / IAP 215

Pluri‐Gaussian Simulation (PGS)

Lithotype rules

G1 G2
© 2013 ‐ IFP Training

Sonatrach / IAP 216
Pluri‐Gaussian Simulation (PGS)

Simulation of two successive episodes

© 2013 ‐ IFP Training
Sedimentation Alteration

Sonatrach / IAP 217

Pluri‐Gaussian Simulation (PGS)

Alterations linked to the lithofacies

© 2013 ‐ IFP Training

Sonatrach / IAP 218
Comparison between conceptual / Digital models

PGS output

© 2013 ‐ IFP Training
Sonatrach / IAP 219

Process‐based model

Process based models mimic physical processes: based on hydrology


equations, with randomized parameters
 Migration
• Point bars (yellow)
• Sand and mudplug (grey and blue green)
 Overbank flood
• Channel lag (orange)
• Levee (dark green)
• Overbank alluvium (bright green)
 Levee breaching
• Crevasse splay I & II, crevasse channels From ENSMP web site
• (tan colors)
© 2013 ‐ IFP Training

 Water table
• Organic rich deposits (purple)
From ENSMP web site

Sonatrach / IAP 220
Contrasted architectures

All simulations are run with the same channel size


Colors darkening with age
 Low aggradation

 Mean aggradation
• In phase system

 Large aggradation
• Quick answer

 Large aggradation

© 2013 ‐ IFP Training
• Preponderance 
of avulsions

Sonatrach / IAP 221

Contrasted architectures

 Parameters can be adjusted to fit stratigraphic sequences defined


by a vertical proportion curve

From ENSMP web site

 Possibility to model meandering 
channelized deposits while 
keeping complexity of sandbodies
arrangements and producing 
contrasted architecture
© 2013 ‐ IFP Training

Colors with age
Sonatrach / IAP 222
Non‐stationary example

 The advantage of TG/PGS on SIS is the possibility to introduce


non stationarity by making the thresholds vary in the space, and
by applying them on stationary Gaussian Functions

© 2013 ‐ IFP Training
Sonatrach / IAP 223

Variation of facies proportions

 Vertically it is always the case, due to the cyclic character of


sedimentation like in particular the variations of the sea level

 Horizontally it is expected when the size of the field is large


compared to the deposition system or to account for transitions
between different systems (marine, fluvio deltaic…)
© 2013 ‐ IFP Training

Sonatrach / IAP 224
Non‐stationary example

 Geologically: trend in the reservoir geology (ex: marine


transgression or regression, compaction…)

© 2013 ‐ IFP Training
Non stationary 
seismic attribute 
Sonatrach / IAP 225

Non‐stationary / Stationary scale – Example

~ Stationary Non stationary 

© 2013 ‐ IFP Training

Seismic attribute  The same seismic attribute…

Sonatrach / IAP 226
Impact of the non stationarity on model

Stationary

Non stationary

© 2013 ‐ IFP Training
Sonatrach / IAP 227

Vertical proportion curves (CPV)

Sand
Shale

© 2013 ‐ IFP Training

Sonatrach / IAP 228
Build a proportion matrix

Sand Non stationary case:
Shale Regionalized proportions

© 2013 ‐ IFP Training
Sonatrach / IAP 229

Build a proportion matrix

Calculation of the proportion matrix
By Proportions (from regionalized proportions) 

© 2013 ‐ IFP Training

Sand
Shale

Sonatrach / IAP 230
Build a proportion matrix

Calculation of the proportion matrix
from Areas

© 2013 ‐ IFP Training
Sand
Shale

Sonatrach / IAP 231

Build a proportion matrix

Calculation of the proportion matrix
By Proportions (from areas proportions) 

© 2013 ‐ IFP Training

Sand
Shale

Sonatrach / IAP 232
Example of results

“Equi‐probable” or
“equi‐possible” 
realizations

© 2013 ‐ IFP Training
Sonatrach / IAP 233

Non stationary example

© 2013 ‐ IFP Training

Turbiditic environment
proximal fan setting

Sonatrach / IAP 234
Non stationary example

Turbiditic environment
proximal fan

© 2013 ‐ IFP Training
Sonatrach / IAP 235

© 2013 ‐ IFP Training

Sonatrach / IAP
Chapter VI:
Uncertainty Quantification

©  2013 ‐ IFP Training
Sonatrach / IAP 237

Monte‐Carlo Approach

 Idea: Generate values from a random process

 How does it work


1) Generate a random p number between 0 and 1
2) Calculating the inverse of the Cumultative Density Function (CDF)
y = F‐1(p)
© 2013 ‐ IFP Training

Sonatrach / IAP 238
Monte Carlo Approach

 It works as soon as you have a statistical distribution of the


variable y

 The CDF F(y) and its inverse F‐1(p) are defined for both continuous
and categorical variables

 The CDF F(y) and its inverse F‐1(p) are defined for both continuous
and categorical variables

 A large number of simulated realizations is required in order to be


consistent

 The random numbers p(i) are generated with an algorithm and a


particular « seed » number

© 2013 ‐ IFP Training
Sonatrach / IAP 239

Monte Carlo Approach: bootstrap

 Example: 17 permeability data are available; f is the


corresponding statistical distribution, F is the CDF

 Methods
1. Draw 17 simulated values from the distribution. This can be seen as
drawing with replacement, some values may be chosen more than
once and other values may never be chosen
2. Calculate the average of the permeability values K and save it as
one possible average
3. Go back to step 1 many time to assess the uncertainty of the mean
© 2013 ‐ IFP Training

Sonatrach / IAP 240
Monte Carlo Approach 

 Easy to generate a lot of a random values

 Work with any kind of variables and distributions

 A main drawback: do not take into account of the spatial


variablility of the variable

© 2013 ‐ IFP Training
Sonatrach / IAP 241

Volumetrics

1. Quantify the uncertainty in gross rock volume by modeling the


surfaces and the fluid contacts stochastically

2. Quantify the uncertainty in net to gross ratio, porosity and


saturation by application of the bootstrap

3. Then perform a bootstrap simulation where gross rock volume,


net‐to‐gross ratio, net porosity are drawn and then multiplied
together to get an oil volume
© 2013 ‐ IFP Training

Sonatrach / IAP 242
Volumetrics

Several simulations of top/bottom, thickness/bottom or 
thickness/top

Simulations of Properties as Porosity, Saturation, Net to gross

© 2013 ‐ IFP Training
Risk curves

Sonatrach / IAP 243

Volumetrics

© 2013 ‐ IFP Training

Sonatrach / IAP 244
Volumetrics

100

90 P90 = 61.01Mm3
80

70
Frequencies

60

50 P50 = 62.51Mm3
40

30

20

10 P10 = 64.31Mm3
0
57 58 59 60 61 62 63 64 65 66 67 68 69

© 2013 ‐ IFP Training
Volumes (Mm3)

Sonatrach / IAP 245

Simulation Optimization

© 2013 ‐ IFP Training

Sonatrach / IAP 246
Simulation Optimization

CDF

1
0.6
Probability of porosity > 10%

0 10 20 Porosity
100
90
80
70
60
50
40
30
Probability maps 20
10
threshold maps 0
 Risk analysis

© 2013 ‐ IFP Training
n equiprobable maps of the same property
(same distribution, same variogram, same
conditioning data)

Sonatrach / IAP 247

Simulation Optimization: Probability

© 2013 ‐ IFP Training

Sonatrach / IAP 248
Simulation Optimization: Quantile

© 2013 ‐ IFP Training
Sonatrach / IAP 249

Simulation Optimization

© 2013 ‐ IFP Training

Sonatrach / IAP 250
Confidence intervals

 In statistics, a confidence interval is an interval estimate of a


population parameter. Instead of estimating the parameter by a
single value, an interval likely to include the parameter is given.

 Different confidence interval levels can be defined: 95%,50% ….

 C = P (u(x)< w < v(x)), with C the confidence level

© 2013 ‐ IFP Training
Sonatrach / IAP 251

Confidence intervals

 We have a sample of 20 values of Porosity (). The distribution is


normal. The sample mean ( X ) is equal to 12%. We compute a
95% confidence interval.
X  
 We define: Z 
 / N

 Z has a zero mean

 We define: P(‐ z < Z < z) = 0.95


© 2013 ‐ IFP Training

Sonatrach / IAP 252
Confidence interval

 Then, we have: (z) = P (Z  z) = 0.975


Z = ‐1((z)) = ‐1(0.975) = 1.96

 At the end:
X  
P ( ‐ z < Z  z) = P (‐1.96  / N  1.96)
 
=P( X  1 .96 *    X  1 .96 * )
N N

© 2013 ‐ IFP Training
Sonatrach / IAP 253

© 2013 ‐ IFP Training

Sonatrach / IAP
Chapter VII: 
Geostatistics in Integrated studies

©  2013 ‐ IFP Training
Sonatrach / IAP 255

Volumetrics

Volume computation

Objective

 Estimate the amount of oil in place


• for the whole reservoir
• for each layer / compartment

 Estimate the associated uncertainties


• Multiple realizations
© 2013 ‐ IFP Training

Sonatrach / IAP 256
Volumetrics

Definitions
 Hydrocarbon present in the reservoir
• f (natural object)
• f(structural model, Fluid contacts, Sw, )

 Original Hydrocarbons In Place (OHIP)


• Hydrocarbons present in the reservoir and considered to be mobile
during production (in the porous volume those whose characteristics
are above cut‐off values)
− f (natural object, cut‐off)
with cut‐off = f (fluid, recovery mechanism, empirical estimation)

© 2013 ‐ IFP Training
OOIP=Original Oil In Place

Sonatrach / IAP 257

Volumetrics

Definitions

 At the field scale

Hydrocarbon Initially In Place (HIIP) is the total volume of


hydrocarbons (gas, oil, dissolved gas) estimated to be contained in
the field.

... refers to Static Conditions, i.e. before producing the wells/field

GIIP = Gas Initially In Place,


OIIP = Oil Initially In Place,
OGIP = Original Gas In Place,
OOIP = Original Oil In Place,
© 2013 ‐ IFP Training

OHIP = Original Hydrocarbons In Place

Sonatrach / IAP 258
Volumetrics

Definitions
Reserves
 Hydrocarbons considered as recoverable during production, for some
development conditions.
• f (Accumulations, macroscopic recovery, economics)
• with Economics = f (Number of wells, production kinematics, recovery
mechanism)

Reserves are a subset of resources...


(SPE/WPC/AAPG)
... refers to Dynamic conditions, i.e. to well/field production

© 2013 ‐ IFP Training
Reserves = HIIP x RF

Sonatrach / IAP 259

Volumetrics

Estimation of OHIP / OOIP


Taking into account
The simplest method all geological trends
and information bias
 OOIP = BRV * N/G * Phi * So * (1/Bo)

• N/G = Average of N/G in all available wells


• Phi = Average of Phi in all available wells
• So = Average of So in all available wells
• 1/Bo = Average of 1/Bo in all available wells
• BRV = Volume below Top Surface – Volume below base surface
© 2013 ‐ IFP Training

Hence, the only parameter to assess is BRV

Sonatrach / IAP 260
Volumetrics

Definitions

 Basics of HIIP volume calculation

OIIP = Oil Initially In Place GIIP = Gas Initially In Place


= =
Gross Rock Volume (GRV) Gross Rock Volume (GRV)
x Net/Gross ratio (N/G) x Net/Gross ratio (N/G)
x Average Porosity () x Average Porosity ()
x Oil Saturation (So) x Gas Saturation (Sg)
x 1/ volume Factor (1/Bo) x 1/ volume Factor (1/Bg)
+ Associated Dissolved Gas + Associated Liquid
=

© 2013 ‐ IFP Training
=
OIIP x Rs GIIP x Yield

Sonatrach / IAP 261

Volumetrics

Estimation of OHIP / OOIP

Be careful about the reliability of these estimates!

 OUR knowledge of BRV, N/G, Phi, Sw is always a result of


extrapolations / interpolations based on INTERPRETATIONS and
CONCEPTUAL MODELS (velocity, dual porosity, deep sea fan,
exponential decline,…)

…WHICH CAN GENERATE THE WORST MISTAKES !!!
© 2013 ‐ IFP Training

Sonatrach / IAP 262
Upscaling

Needs for upscaling

 Upscaling and gridding are not independent:

 What is necessary: a good grid or a good upscaling?

© 2013 ‐ IFP Training
Sonatrach / IAP 263

Upscaling

Needs for upscaling

 The mathematical equations describing the physics are


discretized to be numerically computable (finite difference)
• Properties are uniform over each cell
• Geometry of cells is simple:
− Cartesian
− Corner point

 Artefacts due to discretization and numerical schemes


• Numerical dispersion
• Grid orientation effect
© 2013 ‐ IFP Training

• Change of Kr shape (per rock type)


• Or use permeability multipliers

Sonatrach / IAP 264
Upscaling

Reservoir Grid (coarse grid) definition


 Type of grid
• Cartesian or corner point (to follow simulator grid standards)

 Constraints to consider for horizontal gridding


• Structure, especially faults
• Wells (present, future)
• Best to refine grid when properties change rapidly
• Permeability anisotropy
• Minimum number of cells between wells
• Problem to investigate (e.g., infill drilling)

 Data to define

© 2013 ‐ IFP Training
• Use a specialized gridding software (best, especially for faulted reservoirs)
• Define DX, DY, DZ, Z (top or center) for each cell of a Cartesian grid
• Define X, Y, Z of each cell corner point

Sonatrach / IAP 265

Upscaling

Reservoir Grid (coarse grid) and geological grid

 True box
• Center of fine grid cells in the coarse cell

 Bounding box
• Center of fine grid cells in the bounding box
© 2013 ‐ IFP Training

Sonatrach / IAP 266
Upscaling

Reservoir Grid (coarse grid) and geological grid

Can be resolved with topological approach

© 2013 ‐ IFP Training
Sonatrach / IAP 267

Upscaling

Input Data: geometry definition


 Layering
• Must respect the geological units: best lateral correlation
• The geological layers may be split according to:
− Petrophysical contrasts
− Physics of flow: fluid segregation cannot be modelled with a single layer
(high density contrast, high kv/kh)

 continuous layers of either very high permeability or very low


permeability must be respected:
• A thin very permeable layer must be represented
• A shale barrier is a good place for a layer interface
• Don't aggregate layers with too different k/f
© 2013 ‐ IFP Training

• Aggregate parts of reservoir with close k/f characteristics


• Avoid big thickness contrast between consecutive layers

Sonatrach / IAP 268
Upscaling

Input data of calculators: pre‐processing and fluid flows

The numerical computation is made on 


Input of numerical simulator:
a network 
blocky model
Node properties:
Block properties:
• Porous volume
• Dimension
• Depth of gravity center
• Depth
• Permeability

© 2013 ‐ IFP Training
• Porosity Link properties
• Transmissivity
• Kr (upstream)

Sonatrach / IAP 269

Uncertainties

How to quantify uncertainties?

Deterministic approach Stochastic / probabilistic approach

Identify main uncertainties with 
impacts on HIIP

Develop a case for each  Get probability distribution for 
combination of parameters uncertain parameters of geostatistical
analysis

3 HIIP values to characterize  HIIP characterize with a probability 
the field: distribution: P90, P50, P10
MIN, MED, MAX
© 2013 ‐ IFP Training

Sonatrach / IAP 270
Copyright©2003. ENSPM Formation Industrie - BmD. All rights reserved
Uncertainties

How to quantify uncertainties – Estimation of OHIP / OOIP?

 Deterministic evaluation
• Obtained by combining 2D or 3D grids of the various geometrical and
petrophysical parameters from equation:
− ONE value
− Still a certain culture of determinism in industry...

© 2013 ‐ IFP Training
Sonatrach / IAP 271

Uncertainties

How to quantify uncertainties – Estimation of OHIP / OOIP?

Method of scenarii (deterministic)

 Description
• Optimistic hypotheses for all variables = "maxi" case
• Reasonable hypotheses for all variables = "most likely" case
• Pessimistic hypotheses for all variables = "mini" case
© 2013 ‐ IFP Training

Sonatrach / IAP 272
Uncertainties

How to quantify uncertainties – Estimation of OHIP / OOIP?

Method of scenarii (deterministic)

 Advantages
• Each parameter is controlled and known from prior knowledge
• Well suited for major choices (geol. model, fault pattern)

 Drawbacks
• Difficult to associate probability for each scenario
• Avoid variables which vary continuously in an uncertainty range

© 2013 ‐ IFP Training
Sonatrach / IAP 273

Uncertainties

How to quantify uncertainties – Estimation of OHIP / OOIP?

 Probabilistic evaluation
• Principle: to take into account the uncertainties related to the various
involved parameters
• 1D, 2D, 3D geostatistical approach
© 2013 ‐ IFP Training

Sonatrach / IAP 274
Uncertainties

How to quantify uncertainties – Estimation of OHIP / OOIP?

1D: single points statistics  Scalar Monte Carlo simulation

 Ignoring dependencies among the various input parameters or


between reservoir compartments can lead to under estimation of
the overall uncertainty range

 Important:
• Filling of the reservoir represented only with average values
Less dispersion than punctual values (support effect)
• Chosen formula has to represent the major sources of uncertainty
• Correlations between variables have a strong impact on the results

© 2013 ‐ IFP Training
Sonatrach / IAP 275

Uncertainties

How to quantify uncertainties – Estimation of OHIP / OOIP?

Scalar Monte‐Carlo approach

 Description:
• Reservoir = cube with associated random variable for the volume and
the average parameters
• Variables can be correlated
• Variables are combined for each random draw (crystal ball principle)
© 2013 ‐ IFP Training

Sonatrach / IAP 276
Uncertainties

How to quantify uncertainties – Estimation of OHIP / OOIP?

Scalar Monte‐Carlo approach

 Advantages
• Fast and flexible to assess reasonable Q10 and Q90

 Drawbacks
• Difficult to quantify uncertainty on BRV
• No representation of heterogeneities

© 2013 ‐ IFP Training
Sonatrach / IAP 277

Uncertainties

How to quantify uncertainties – Estimation of OHIP / OOIP?

 Scalar Monte‐Carlo approach


GRV N/G  Sw

STOOIP

Random number generation

GRV x N/G x  x (1-Sw)


© 2013 ‐ IFP Training

STOOIP = -----------------------------
Bo

Sonatrach / IAP 278
Uncertainties

How to quantify uncertainties – Estimation of OHIP / OOIP?

 Scalar Monte‐Carlo approach

probability
most likely
1
0.9

0.5
mode

median
mean

© 2013 ‐ IFP Training
0.1
reserves
25 50 (MMBOE)
proven probable possible

Sonatrach / IAP 279

Uncertainties

How to quantify uncertainties – Estimation of OHIP / OOIP?

Pitfalls in Monte‐Carlo simulations

 Selecting a distribution for a parameter is often arbitrary

 It is important to have in mind the “support effect” when the


range of variation of a parameter is given
It is necessary to have a clear view of the geological model
© 2013 ‐ IFP Training

Sonatrach / IAP 280
Uncertainties

How to quantify uncertainties – Estimation of OHIP / OOIP?

 Pitfalls in Monte‐Carlo simulations


• Parameters are strongly correlated (F and So for ex.), which has to be
taken into account during simulations
frequency

frequency
ρ=+1 ρ=0
V1 V1
frequency

frequency frequency
frequency

V2 V2

© 2013 ‐ IFP Training
fromJ.Pouzet
F(V1,V2) F(V1,V2)

Sonatrach / IAP 281

Uncertainties

How to quantify uncertainties – Estimation of OHIP / OOIP?

2D  Mapping Monte Carlo method

 The spatial variability in structural and petrophysical parameters


can be modeled
• The definition of GRV is an important contributor to overall
uncertainties
• Structural uncertainties are modeled by geostatistical tools
© 2013 ‐ IFP Training

Sonatrach / IAP 282
Uncertainties

How to quantify uncertainties – Estimation of OHIP / OOIP?


3D: stochastic modeling of heterogeneities
 Requires a description of the spatial distribution, size, and/or
correlation length for each of the sedimentary blocks involves the
following steps:
• Definition of a conceptual geological model and framework
• Classification of well data into appropriate geological classes (facies)
• Characterization of inter‐well properties defining uncertainties in the
data
• Generation of 3D heterogeneity models (distribution of facies or
reservoir parameters)
• Calculation of multiple, equiprobable realizations that can be ranked

© 2013 ‐ IFP Training
and integrated with other reservoir technical uncertainties

Sonatrach / IAP 283

Uncertainties

How to quantify uncertainties – Estimation of OHIP / OOIP?

 Issues with geostatistical approach


• Are all the realizations realistic?
• Are they all compatible with production data?
• How many realizations are required to obtain a statistically
consistent estimation of the mean and variance of the result
calculated from the realizations?
© 2013 ‐ IFP Training

Sonatrach / IAP 284
Statistics and geostatistics

Key points to keep in mind

 For Reservoir characterization step


• Use statistics
− As a first level of data analysis
• Use geostatistics:
− To improve understanding regarding rocks and petrophysical
distribution
• Deliverable:
− Improve conceptual models (e.g. Sedimentologic, Stratigraphic,
Diagenesis models)
− Parameters for modeling (e.g. variance, Range and sill, nugget
effect)
 For modeling step
• Use geostatistics
− Populate a realistic data distribution between well integrating

© 2013 ‐ IFP Training
constraints as:
» Seismic map
» Sedimentological model

Sonatrach / IAP 285
 

You might also like