Analyse

GEOSTATISTIC 15 3 the data

Sand
Un
known

Defind a

© 2006 ENSPM Formation Industrie - IFP Training

representative
Populate
value
a volume
GRE 2006
 Classe 1
223
Sand 1,54
Shale Rouge

What is a DATA
13 Levee Vert Classe 2

© 2006 ENSPM Formation Industrie - IFP Training

1432
Understand Dolomite
0,456 the data
GRE 2006
 Data:
Continuous: variable: porosity, thickness, time,
Real; 1,55… 1,56….2204… 245…- 45… +345
Discrete: sand, shale, channel, levee
Nominal; 1= Sand, 3= Levee, 1,0, 1

Porosity log Facies log

© 2006 ENSPM Formation Industrie - IFP Training

GRE 2006
 Stationnarity
Elevations, porosity

Probabilitic model
The variance does not depend on the location
1st order arithmetic mean is constant
Stationarity depends on the scale
Window observation
Non Stationnarity
Beware of the sampling
Facies changes, compartmentalization

Regional trends can be detected in the data

trends

© 2006 ENSPM Formation Industrie - IFP Training

GRE 2006
 WORK FLOW
Set of data
Analysis as univariate outcome
Clean the data
Check the validity
of the data Analysis as multivariate outcome

Evaluate the spatial distribution

© 2006 ENSPM Formation Industrie - IFP Training

Condensate the information
Defined the geostatistical approach
Extrapolation, Modelisation, Prediction
GRE 2006
 Univariate Data analysis

Distribution fonction:
Modeled with: mean, max, variance

Ramdon variable
5% levee, 45% shale plug, 50% sand
Take only certain values with certain probabilities

Sand

© 2006 ENSPM Formation Industrie - IFP Training

Shale
Sand

1,54 4,5 8,25

GRE 2006
 Analyse the data

Histogram Cumulative
Histogram
pdf distribution
cdf
Probability distribution fonction Cumulative distribution fonction

Assymetry
Limits Shape
Variation coefficient
Bornée
Sign M-
M-m
Sign M-
M-m

Central value
Position value Dispersion
Median value

© 2006 ENSPM Formation Industrie - IFP Training

Quantile Variance
Mode value
P10, P50, P90 Standard deviation
Geometric mean
Coefficient of variation
Harmonic mean
Arithmetic mean
GRE 2006
 Basic Data analysis
Histogram distribution
Mean; maximum, minimum, mode
Variability, dispersion of data ,variance, standard deviation
Coefficient of variation: Is volume dependant
Cumulative histogram
Grade or cell size dependent
Covariance gives the joined variations of X and Y around their means, basic
tool to measure the relationship between two random variables

Probability density function; P10, P50, P90

© 2006 ENSPM Formation Industrie - IFP Training

Correlation and regression line, Linear, Exponential…

GRE 2006
 Central values

Mean= ∑ φ ( xi ) E(x)
1 N
N 1 Harmonic mean
N
Geometric mean mhz = N
1
(∑ )
mgz = N ∏ iN=1 z ( xi ) i =1 z ( xi )

Mode: Most likely value

Median: Value for which 50% values above 50 %

© 2006 ENSPM Formation Industrie - IFP Training

m

Lognormal Multi modal Cumulative histogram

GRE 2006
 Dispersion measurement
1 N
Variance= σ = N ∑
2
2
[φ ( xi
) − m ]
1

Standard deviation= σ 2= σ
Is volume dependant, sampling

Coefficient of variation= σ /m

Position value +/- σ

+/-2 σ
68.27%are between
95,45%are between

quantile=

© 2006 ENSPM Formation Industrie - IFP Training

P90 Q25, Q50 (median), Q75

P10
Q 10, Q90

GRE 2006
 Shape of the distribution
m-M>0 m-M=0 m-M<0
Signe of M-M

M-m m=M m-M

Coefficient de variation
σ m CV=Small CV=large

0,15 0,20

Truncated distribution, and Untruncated

© 2006 ENSPM Formation Industrie - IFP Training

CV=large

Triangular Uniform Normal Lognormal

GRE 2006
 Multivariate Data analysis

Linear regression
Covariance =

C(X;Y) = 1/n Σ (xi-

(xi-mx)(yi
mx)(yi-
)(yi-my)
my)

Variable Y
Ex: X=aY+b
Strong correlation if C(X;Y) is>0 or <0
If C(X;Y)=0 no correlation
Log K=alog phi+b

Correlation coefficient = gives the linear

trend of the relationship between X&Y
Variable X

ρ(X,Y)=
-0,8
Y Y
el el Y

© 2006 ENSPM Formation Industrie - IFP Training

ba ba el
ir ir ba
a a ir
V V a
V
Variable X Variable X Variable X
ρ(X,Y)= 0 ρ(X,Y)= C(X,Y) /σ
/σx σy
ρ(X,Y)= +0,8
GRE 2006
GRE 2006
value
What is the
representative

© 2006 ENSPM Formation Industrie - IFP Training

 Question : What is the representative value for this cell

Problem:
Integration of the informations
Capture the variability
Representativity
Pick a value reproducing the geological information
?

© 2006 ENSPM Formation Industrie - IFP Training

GRE 2006
 Most of the linear combinations need additives data

10% 1500
md

5% 30md

Porosity Permeability

© 2006 ENSPM Formation Industrie - IFP Training

10+5= 15/2= 7,5% 1500+30= 1530/2= 765k
Arithmetic mean 2/K= 1/1500+1/30

GRE 2006
2/K= 51/1500 = 58,8k
 UPSCALING

Fine Grid (geostatistical model) Coarse Grid (reservoir model)

© 2006 ENSPM Formation Industrie - IFP Training

Capture the variability
Capture the heterogeneity

GRE 2006

Goal: find the best possible values to put into the simulator to modelize the flow.
Depends on local conditions (pressure gradient, vicinity of wells).

A compromize between the physical upscaling and the need for limited
amount of cells (geostatistical models often have million to tens of million
of cells while reservoir models have tens of thousands to hundred of
thousands of cells )
Upscaling
static ( k, PHI)
dynamic ( Kr, Pc )

© 2006 ENSPM Formation Industrie - IFP Training

– Reduce the number of layers
– Reproduces saturation distribution in near-well regions

GRE 2006
 UPSCALING

© 2006 ENSPM Formation Industrie - IFP Training

Integrate the
spacial variability
of the data
POPULATING A GRID
GRE 2006
 We have to
quantify the
spacial
continuities of
the variable

© 2006 ENSPM Formation Industrie - IFP Training

The spacial correlation and the covariance will decrease
GRE 2006
when the distance increase
 18 14 14 15
12 12 16 13
14 20 13

16 17

?
12
Geological
13 15 14 18
17
3d 20
18
14 18
4d
12
15 16 15
18
map
18
12 Y 13
14
Spacial
17
X 20
13
15 15
20
17
12
12
20
16
Distribution
13 14
of the data
12 12 18 16
13 17 14
16 13 15 14

Lag distance

© 2006 ENSPM Formation Industrie - IFP Training

d
2d
3d
Var=1/N(1/2
Var=1/N(1/2 (X-
(X-Y)2)
4d
GRE 2006
 The correlation between two sets of data could be
dependant of the distance V(Xi+h)
V(Xi+h)
(more closer there are more dependent there should be) Variance

V(Xi)
V(Xi)
Is the mean of the square root of the differences
between a value at location 0 and 0+1.
1/N (1/2 Sum ( Xi-X(i+d)) 2)

Sill: Noisy shape: anisotropy; if no trend in the data this is simply equal to
variance of the data 4

Semi variogram: Sill 3

Nugget: Noise, nugget effects Range
2

© 2006 ENSPM Formation Industrie - IFP Training

Range: Correlation length No correlations

1
Nugget
Shape:
1 2 3 4 5 6
Lag distance h

GRE 2006
 0.0 5.0 10.0 15.0

15.0 15.0

Colors

2.4
10.0 10.0
1.8

1.2

0.6

5.0 5.0 -0.6

-1.2

-1.8

-2.4

-3
0.0 0.0
0.0 5.0 10.0 15.0
Test

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

1.00 1.00

2
–  --- 
h
1 h 3 
γ ( h ) = C ( O ) -- ---  – --  --  
3 h
γ ( h) = C ( O )  1 – e
0.75 0.75 a 
2 a 2 a  
a = Range a = Range
0.50 0.50

Practical Range = a Practical Range = 1.73a

0.25 0.25

0.00 0.00
0. 1. 2. 3. 4. 5. 6. 7. 8.

Variogram porosity (omni) - Mod

linear model

Variogram

© 2006 ENSPM Formation Industrie - IFP Training

gaussian model
models
spherical model
exponential model
GRE 2006
 Beware:
The values evolve systematically in one direction (trend)
The values can be approach with a constant (stationarity)
What is the scale of your observation

Cte or fct

trend

© 2006 ENSPM Formation Industrie - IFP Training

scale
GRE 2006
 Dealing with Trends
Trends leads to parabolic shape
Remove trends from data
Compute variogram on these data
Simulate residuals everywhere
Add trend at the end
Prescence of
trend

© 2006 ENSPM Formation Industrie - IFP Training

Linear trend

Cyclicity in the data

GRE 2006
 Goals: Extend the sampled data to the whole field

Deterministic: 22,5/7,5= 15
Estimation
Kriging: Exact interpolator

7,5
Simulation
? 22,5

Stochastic: Probability of fonction

Pixed based
Object based

© 2006 ENSPM Formation Industrie - IFP Training

Need a good understanding of the data
A representativity of the geological geometry with the heterogeneities
Correlable with dynamic data, pressure, production
GRE 2006
 IMPACT DU MODELE GEOLOGIQUE
2 concepts, testés sur un niveau réservoir
Tidal channels Coastal bars

© 2006 ENSPM Formation Industrie - IFP Training

0.06 0.
2 images équiprobables, Phi 2
Accumulations différentes ?

GRE 2006 Copyright©2003. ENSPM Formation Industrie - BmD. All rights reserved
 Geostatistics do not invent a model for you.

Continuous values Discrete values

Distribute heterogeneities

Estimation Simulation Simulation

Scenarii
Ramdon fct: mean; variogram & covariance
Boolean
Deterministic Geostatistic
Stochastic pixel
Variogram

Moving average Kriging Sequential O O O

Object
O O O
Polygone of influence Optimal estimate Gaussian O O O
Perform a linear estimation
Inverse of distance …

© 2006 ENSPM Formation Industrie - IFP Training

Collocated kriging
Created an alternative image with
Created an estimation equal probability of existence
Conventional geostatistics Account for the degree of uncertainties

GRE 2006
 Deterministic: linear estimate

Mean, Moving average

Polygone of influence,
Inverse of distance

Linear interpolation

© 2006 ENSPM Formation Industrie - IFP Training

Inverse of distance
Polygone of influence
GRE 2006 After Beicip Franlab
 Kriging: Variogram will be a crucial function (sill, range, nugget and shape)
Honored the data, is an exact interpolator (if no nugget)
Integrate distance between points
Kriging tends to converge to the mean
Produced smoothed map
Geometry of the hard data
Best guess
Do not reproduced heterogeneities, variability of the data
The varogram is knowned and stable in all the area

Cokriging: estimation using secondary data (local or full)

© 2006 ENSPM Formation Industrie - IFP Training

Seismic amplitude map with porosity map

NB: Declustering: Wi= 1/Nlo*Lo

GRE 2006
 Kriging is a process to produce a linear estimate with a minimum error variance

Simple kriging
Requires the knownledge of the mean . A trend is introduced

The variogram is presumed to be known and stationary over the domain

Ordinary kriging
The mean is constant but unknown

Universal kriging
Ecart à la moyenne are used in the kriging

Universal kriging with external secondary (soft) data

© 2006 ENSPM Formation Industrie - IFP Training

An external map is introduced

Cokriging or localised Cokriging

An defined variable is introduced and correlated
with the variable to estimate
GRE 2006
 Stochastic approach “Partially random with some predictability”

Drawn one or several realisations of the random fonction

Generate multiple realizations which honor hard data
Is not a prediction, is a probabilistic view not “limited”
Integration of realistic level of uncertainties
Can be used to perform risk analysis
Hard data from control points are not sufficient to describe the reservoir

© 2006 ENSPM Formation Industrie - IFP Training

GRE 2006
 Conditional simulation kriging

output “equiprobable” realizations one deterministic model

“honors wells honors wells

properties
honors histograms, variograms minimize average error

image “noisy, smooth

same variability everywhere

use flow simulations, mapping

uncertainty computations volumetrics

© 2006 ENSPM Formation Industrie - IFP Training

Kriging and Simulation

GRE 2006
 Kriging versus Simulation
0. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.
0. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.
10. 10.
10. 10.

9. 9.
9. 9.

8. 8.
8. 8.

7. 7. Colors
7. 7. Colors

6. 6. 12
6. 6. 13
11.4
5. 5. 10.8

10.2
Kriged Porosity 5. 5.
12.1

11.2
4. 4. 10.3
9.6
4. 4.
9
9.4
3. 3.
8.4 8.5
3. 3.
7.6
2. 2. 7.8

7.2 2. 2. 6.7

1. 1. 6.6 5.8
1. 1. 4.9
6
0. 0.
4
0. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.
Test 0. 0.
0. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.
Poro krig 38 wells Test

0. 1. 2. 3.
Sim_#0001
4. 5. 6. 7. 8. 9. 10. Same Histogram
Same Variogram
10. 10.
0. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

10. 10.
9. 9.

9. 9. 8. 8.

8. 8. 7. 7. Colors

7. 7. Colors 6. 6. 13

12.1

13 5. 5. 11.2
6. 6.

© 2006 ENSPM Formation Industrie - IFP Training

Simulated 5. 5.
12.1

11.2
4. 4.
10.3

9.4

Porosity 4. 4.
10.3

9.4
3. 3.
8.5

7.6

with the same 3. 3.

8.5

7.6
2. 2. 6.7

5.8
1. 1. 4.9

Variogram 2. 2. 6.7

5.8 0. 0.
4

0. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.
1. 1. Test
4.9
Sim_#0002

Equiprobable realisation
4
0. 0.
0. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.
Test

Sim_#0001

GRE 2006
 Conditional simulations properties

•Each conditional simulation is equally probable.

• Each conditional simulation must have the same variogram and histogram as the
conditioning data and must honor the data values.

• The anamorphosis allows one to normalize the data and to back-transform the
conditional simulations.

• The mean of several conditional simulations tends towards the kriging estimate.

© 2006 ENSPM Formation Industrie - IFP Training

• The dispersion variance of several conditional simulations tends towards the
kriging estimation variance. The mean of these conditional simulations is…kriging.

GRE 2006
 Simulation method
Continuous/Discrete variables
porosity, permeability, depth, thickness, impedance…
facies, geological objects...
Pixel method
Generation of objects, geological or structurals
Sequential Gaussian Simulation (SGS), for continous data

Sequential indicator Simulation (SIS),

estimate the probability that the point is sand, shale: p-1
Truncated gaussian Method
Object method

© 2006 ENSPM Formation Industrie - IFP Training

Generation of succession of facies or petrophysical properties

Probabilistic model take in account the uncertainties

GRE 2006
 Pixel based
(Ex: Carbonate reservoir,
petrophysical properties)

Object based

© 2006 ENSPM Formation Industrie - IFP Training

(Ex: channels sands,
fractured reservoirs)

GRE 2006
 Pixel method

• No direct “mapping” of the heterogeneities

•chronostratigraphy

•Petrophysical properties

© 2006 ENSPM Formation Industrie - IFP Training

GRE 2006
 Object method
• Clear understanding of the geometry of the objects

© 2006 ENSPM Formation Industrie - IFP Training

• Create objects

GRE 2006
 Cube de
proportion

por sdt
m por sdt
por dolo

© 2006 ENSPM Formation Industrie - IFP Training

por calc
m por dolo
low por calc
barrier
GRE 2006
 ref=top

prop

© 2006 ENSPM Formation Industrie - IFP Training

ref=bot

Impact of Geological Model and Reference Level

GRE 2006
Not all stochastic modeling studies need a Cadillac technique; some
studies will do fine with the Volkswagen technique or with a solid
pair of rollerskates…
(Srivastava, 1996)

Stochastic models for: assessing the impact of uncertainty

Monte Carlo risk analysis
honoring heterogeneity and complex information

© 2006 ENSPM Formation Industrie - IFP Training

GRE 2006
 Database Quality Control

Log Fluids Studies Seismic

Sedimentology
Interpretation (Well test, PVT) Interpretation

Data Integration

Building of the Geological Model

Deterministic Stochastic

© 2005 ENSPM Formation Industrie - IFP Training

Oil in place Computation

Upscaling

Production model
Géostatistiques Pétrolières
Brigitte DOLIGEZ - IFP - September 2005 24 - 28 Septembre 2005 - Boumerdes 140

Reservoir modeling = Need to quantify

- quantitative understanding: sedimentary environment, facies variations, evolution of properties…
must be converted into quantitative data: facies percentages, petrophysical characteristics…
- to generate information in all points of a given space and to quantify the impact of hypotheses on
valuesare heterogeneous; their heterogeneity is correlated at a reservoir scale; the heterogeneity
- reservoirs
will influence recovery

© 2006 ENSPM Formation Industrie - IFP Training

- A reservoir is the product of 3 types of geological events: sedimentary process; diagenetic evolution;
mechanical stress
These 3 types of events induce 3 types of heterogeneities: sedimentary; diagenetic; tectonic

GRE 2006
 Workflow
• Data Analysis

• Simulation of Lithofacies

• Simulation in Petrophysical Properties

• Petrophysical Model Upscaling

© 2006 ENSPM Formation Industrie - IFP Training

GRE 2006
 • S t a t ic :
– T h e c h r o n o lo g ic r e c o n s t r u c t io n o f t h e s e d im e n t a r y p r o c e s s a llo w s
t o p r e d ic t t h e s p a t ia l o r g a n is a t io n o f t h e h e t e r o g e n e it ie s
( s e q u e n c e s t r a t ig r a p h y )

• D y n a m ic :
– T h e r e c o n s t r u c t io n o f t h e b e h a v io r o f t h e r e s e r v o ir a llo w s t o
u n d e r s t a n d a n d t o p r e d ic t t h e s p a t ia l o r g a n is a t io n o f t h e
h e t e r o g e n e it ie s f o r f lu id f lo w

2005 ENSPM Formation Industrie - IFP Training

A scale problem:

Scale of geological phenomena: scale of heterogeneities effectively present in the reservoir

© 2006 ENSPM Formation Industrie - IFP Training

Scale of available data: scale of volumes which can be described or measured

Scale of tools: scale of volumes actually explored by the tools

GRE 2006
 What geostatistical modeling is

• Tool box generating realistic 3D representations of

heterogeneities
• Generate equiprobable realizations of the subsurface
i.e all compatible with the data used in input to the model

• Variability between geostatistical realization =

Measure of the uncertainty remaining
After constraining the realizations by the input data

© 2006 ENSPM Formation Industrie - IFP Training

• Treats known information as deterministic
and unknown information as probabilistic.

From P.Lamy
GRE 2006
 What geostatistical modeling is not

• Is not a substitute for deterministic approaches

• Is not a substitute for ignorance about geology of a field
• Is not a “coin-tossing” algorigtm to decide which facies are
present between wells.

• It is not fully objective

© 2006 ENSPM Formation Industrie - IFP Training

since some of the input parameters may be wrong

From P.Lamy
GRE 2006
 Geostatistics

Quantify the geographic variability of the data/

Variogram
Produce an estimate : kriging and cokriging

Dispersion of the heterogeneities : Simulation

© 2006 ENSPM Formation Industrie - IFP Training

Take in account the geographical location

From P.Lamy

GRE 2006
 EXERCISES

© 2006 ENSPM Formation Industrie - IFP Training

BMD
GRE 2006 Copyright©2003. ENSPM Formation Industrie - BmD. All rights reserved