14 SEISINV Course2021 Application of MultiPoint GeoStatistics

MULTIPLE-POINT GEOSTATISTICS (MPS)
USING DIRECT SAMPLING ALGORITHM
Sonny Winardhi
22/11/2019
Geo-Lunch Talk
Sharing and Discussion Forum - Prodi Teknik Geofisika
Geo-Lunch Talk
TG - Sharing and Discussion Forum
Introduction
Key Words:
Geo-statistical
Histogram, estimation
Geo-statistics (Kriging)
pdf, cdf
Spatial Pattern Variogram Geo-statistical
Estimation and Two-point Simulation
Simulation Statistics
Sequential
Multi-point Gaussian
Statistics Simulation
Direct (SGS)
SNESIM
Sampling
Reservoir
Distribution
Geo-Lunch Talk
Traditional/Classical Geo-Statistic
N(m,s)
Traditional Geo-Statistic
(two-point statistics)
Estimation
Geo-Lunch Talk
N(m,s)
Simulation
Geo-Lunch Talk
Simulation
Geo-Lunch Talk
Limitation of Traditional Geo-Statistic
1 2 3
Variograms EW Variograms NS
3
1.2 1 1.2 2
2 1
0.8 0.8
3
0.4 0.4
0 10 20 30 40 0 10 20 30 40
Variogram is not enough to characterize connectivity pattern.

A prior geological knowledge/interpretation is required.
Geo-Lunch Talk
Limitation of Traditional Geo-Statistic
A geologist’s
interpretation of
isopach data.
Meandering
channel In-fill channel
Geological knowledge
cannot be integrated into
Traditional Geo-
Statistics to create four
different maps
(representing four
different geological
Paleo-valleys
Barrier bar and environments).
tidal channel
(Chiles and Delfiner, 1999)
Geo-Lunch Talk
Classical Geo-Statistic vs MPS
Traditional Two-point Geostatistics
B1
A
B2
B3
B={B1,B2,B3}
P(A|B) = N(m,s)
m,s given by Kriging, depend on

Variogram function
Geo-Lunch Talk
Traditional Geo-Statistic vs MPS
Multiple-Point GeoStatistics
Reservoir
= well data
multiple-point
data event
Sequential A
simulation
P(A|B)?
B
Geo-Lunch Talk
Multi-point GeoStatistics
The essential data integration problem…
A geologist believes based on geological data that there is 80%

chance of having a channel at location X
P(A|B)
A geophysicist believes based on geophysical data that there is

75% chance of having a channel at location X P(A|C)
What is the probability of having a channel at X ?
P(A|B,C)
Simple Question, difficult problem…

Geo-Lunch Talk
The essential data integration problem…

A geologist believes based on geological data that there is 80%
chance of having a channel at location X
P(A|B)
A geophysicist believes based on geophysical data that there is

75% chance of having a channel at location X P(A|C)
A petroleum engineer believes based on engineering data that

there is 85% chance of having a channel at location X P(A|D)
What is the probability of having a channel at X ?

P(A|B,C,D)?
Simple Question, more difficult problem…
Geo-Lunch Talk
Bayes Rule
P(A|B)?
P(A|B) = P(B|A).P(A) / P(B)
P(B) = P(B|A).P(A) + P(B|A’).P(A')

Geo-Lunch Talk
Bayes Rule
P(B,C|A) = P(B|A).P(C|A)
P(A|B,C) = P(A). P(B|A).P(C|A) / P(B,C)
It requires Markov Chain Monte Carlo

(MCMC) to eliminate P(B,C)
Geo-Lunch Talk
Simple question, difficult problem…
P(A|B,C,D)?
Recursive Application of Bayes Rule
P(A|B,C,D) = P(D|A,B,C).P(C|A,B).P(B|A).P(A) / P(B,C,D)
P(D|A,B,C) = ?, P(C|A,B)=?, P(B,C,D)=? …

Geo-Lunch Talk
MPS using SNESIM Algorithm
Training image module =

standardized analog model
quantifying geo-patterns
SNESIM algorithm
Recognizing P(A|B) for all possible A,B
Geo-Lunch Talk
MPS using SNESIM Algorithm
SNESIM algorithm
Recognizing P(A|B) for all possible A,B
Memory demanding !!
Multiple-point patterns with adjacent grid nodes

Geo-Lunch Talk
MPS using Direct Sampling Algorithm
How it works?
Simulation neighborhood Training image
Matches !
Copy the value Doesn’t match

Doesn’t match
As soon as one good configuration is found, the value at the

central pixel is assigned to the simulated node.
(Mariethoz et al., 2010)

Geo-Lunch Talk
MPS using Direct Sampling Algorithm
How it works?
Data event constituted of the n closest nodes. Adapts
automatically with the density of informed grid nodes.
(Mariethoz et al., 2010)

Geo-Lunch Talk
Classical Geo-Statistic vs MPS-DS
Direct Sampling
Classical GeoStat
Delta of the Lena River, Landsat (USGS)

Geo-Lunch Talk
Classical Geo-Statistic vs MPS-DS
Direct Sampling
Classical GeoStat
Similar variogram, different

image pattern
Geo-Lunch Talk
MPS-DS of Synthetic Data with constraint
Geo-Lunch Talk
Application of MPS-DS to Field Data
West East
Geo-Lunch Talk
West East
Tidal
Sand
Sand
Shale
Acoustic Impedance
Geo-Lunch Talk
West East
Tidal
? ? ? ? ? ? Sand
Sand
Shale
Acoustic Impedance
Geo-Lunch Talk
Sand
Analog Model
Shale
Acoustic Impedance
Data Integration: Well, Geological

Analog Model, and Acoustic Impedance
resulting from Seismic Inversion.
Training Image
Geo-Lunch Talk

Analog Model, and Acoustic Impedance 1st Scenario
TI
Three Simulations resulting from

using TI, Well, and 50% AI.
AI
Geo-Lunch Talk

Analog Model, and Acoustic Impedance 2nd Scenario
TI
Three Simulations resulting from

AI
Geo-Lunch Talk
Concluding Remarks
• MPS: a way of incorporating prior geological knowledge

into the geo-statistical algorithm.
• MPS: provides a more realistic data integration
approach than traditional Bayesian methods.
• MPS-DS: large flexibility of prior models.

• MPS-DS: a fast and robust sampling of the prior.
• MPS-DS: can be extended to a generic inverse
solution method that honor prior information.
Geo-Lunch Talk
One Realization resulting from

Geo-Lunch Talk
Shale Sand
Geo-Lunch Talk
North South
Geo-Lunch Talk
Geo-Lunch Talk
Synthetic Data Experiment
Discrete Data
Discrete Data TI
Resembles seismic data
Continuous Data
Geo-Lunch Talk
Synthetic Data Experiment: Unconstrained-MPS

Geo-Lunch Talk
Synthetic Data Experiment:

Constrained-MPS
Geo-Lunch Talk
Result of Synthetic Data Experiment
Desired Output
 
Geo-Lunch Talk
Application to Field Data
Seismic as Constraint
Geo-Lunch Talk
Application to Field Data:

Result of MPS and Constrained-MPS
10% Weight of Seismic

will be P(90)

will be P(50)

will be P(10)

14 SEISINV Course2021 Application of MultiPoint GeoStatistics

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

14 SEISINV Course2021 Application of MultiPoint GeoStatistics

Uploaded by

Copyright:

Available Formats

MULTIPLE-POINT GEOSTATISTICS (MPS)

USING DIRECT SAMPLING ALGORITHM

Variogram is not enough to characterize connectivity pattern.

Traditional Two-point Geostatistics

m,s given by Kriging, depend on

The essential data integration problem…

A geologist believes based on geological data that there is 80%

A geophysicist believes based on geophysical data that there is

What is the probability of having a channel at X ?

Simple Question, difficult problem…

The essential data integration problem…

A geophysicist believes based on geophysical data that there is

A petroleum engineer believes based on engineering data that

What is the probability of having a channel at X ?

P(A|B) = P(B|A).P(A) / P(B)

P(B) = P(B|A).P(A) + P(B|A’).P(A')

P(A|B,C) = P(A). P(B|A).P(C|A) / P(B,C)

It requires Markov Chain Monte Carlo

Recursive Application of Bayes Rule

P(A|B,C,D) = P(D|A,B,C).P(C|A,B).P(B|A).P(A) / P(B,C,D)

P(D|A,B,C) = ?, P(C|A,B)=?, P(B,C,D)=? …

Training image module =

Multiple-point patterns with adjacent grid nodes

Simulation neighborhood Training image

Copy the value Doesn’t match

As soon as one good configuration is found, the value at the

(Mariethoz et al., 2010)

(Mariethoz et al., 2010)

Delta of the Lena River, Landsat (USGS)

Similar variogram, different

Data Integration: Well, Geological

Data Integration: Well, Geological

Three Simulations resulting from

Data Integration: Well, Geological

Three Simulations resulting from

• MPS: a way of incorporating prior geological knowledge

• MPS-DS: large flexibility of prior models.

One Realization resulting from

Synthetic Data Experiment

Resembles seismic data

Synthetic Data Experiment: Unconstrained-MPS

Synthetic Data Experiment:

Result of Synthetic Data Experiment

Application to Field Data

Application to Field Data:

10% Weight of Seismic

50% Weight of Seismic

90% Weight of Seismic

You might also like