Information Entropy
Monte Carlo Simulation
March, 27th, 2009
Ayato Kato
Rock Physics
Outline
Shannon’s Information Entropy
Monte Carlo Simulation
Rock Physics 2
Which attribute(s) should We Use ?
Innumerable Seismic Attributes
Type Seismic
Attribute
Major Geological significance
For the purposes of
Conventional
property
Amplitude Lithological
continuity
contrast, bedding
predicting porosity, which
Interval velocity
Acoustic
Lithology, Porosity, Fluid Content
Lithology, Porosity, Fluid Content attributes should we use?
impedance
Volume-related Reflection Reservoir Architecture,
attribute geometry Sedimentary Structure
(Multi-trace attribute) Trace continuity Fault geometry, Fault distribution,
Stratigraphic continuity
- AVO P0 & G
Time curvature, Detailed Reservoir Architecture,
Dip, Azimuth Fault geometry, Fault distribution,
- AI & EI
Pre-stack attribute AVO
Fracture density
Fluid Content, Lithology, Porosity
- λρ & μρ
Impedance
(elastic/S-wave)
- Vp/Vs etc
Poisson's ratio
λ, μ
(Lame constant)
AVOZ Fracture Orientation, Fracture Density, Shannon’s Information
Fluid Content
Instantaneous
attribute
Instantaneous
phase
Bedding continuity Entropy can give us the
Instantaneous
Frequency
Bed thickness, lithologic contrast, fluid
content
solution quantitatively.
Miscellaneous Frequency Fluid content
attribute Attenuation
Anything ???
computed from
seismic traces Rock Physics 3
Shannon’s Information Theory
Shannon and Weaver (1949)
“The Mathematical Theory of Communication”
Defined Quantity of Information
Information Content
Dr. Claude Shannon
Information Entropy
Mutual Information
Rock Physics 4
Information Content
It will snow !!
Houston Low Probability
Very rare (1% ) Very surprised !!
(valuable information) Isnow = 6.64
Calgary High Probability
Very often (70%) Not surprised
(Not-valuable information) Isnow = 0.51
Information Content (Quantity of information)
I = − log P How surprised one would be if the event
happened.
* Base = 2
Rock Physics 5
Information Entropy
Expected Value of Information Content
Information Entropy X = {x1 ,⋅ ⋅ ⋅ ⋅ ⋅⋅, x N }
n
H ( X ) = −∑ Pi ⋅ log Pi • Expected Surprise
• Quantity of uncertainty associated with P
i
Xi Pi (%)
Houston Sunny 33
Cloudy 33
H=1.65 Rain 33
Snow 1
Calgary Xi Pi (%) NY Xi Pi (%)
Sunny 25
Sunny 10
H=1.36 Cloudy 10 H=2.00 Cloudy 25
Rain 25
Forecast is easy Rain 10 Forecast is difficult
Uncertainty is small Snow 70 Uncertainty is large Snow 25
Rock Physics 6
Conditional Entropy
Conditional Information Entropy
H ( X | A) Conditional Information Entropy at a given additional
information (A)
e.g. A = atmospheric pressure change data
NY Weather Pi (%) At given additional Weather Pi (%)
Sunny 25 information A Sunny 50
H(X)=2.00 Cloudy 25
H(X|A)=1.54 Cloudy 25
Rain 25
Uncertainty is large Uncertainties reduce
Rain 10
Snow 25 Snow 5
Mutual Information
I ( X | A) = H ( X ) − H ( X | A)
Quantity of uncertainty reduced by
additional information A
Rock Physics 7
Our Case : Porosity Prediction
Porosity Prediction by Seismic Attribute
X : Porosity (continuous variable)
A : Seismic Attribute
H(X) : Information Entropy for Porosity’s PDF
H(X|A) : Information Entropy for Porosity’s PDF at given seismic attribute
PDFs
H(φ|Vp,Vs)=2.87
Adding more seismic attributes
H(φ|Vp)=3.06 PDF shape → Narrow, steep
Uncertainty → Decrease
In. Entropy → Decrease
H(φ)=3.44
Mukerji et al. (2001) Rock Physics 8
Information carried by Seismic Attributes
For example,
X = Porosity
A = Seismic Attribute
Mutual Information
I ( X | A) = H ( X ) − H ( X | A)
Information entropy Conditional entropy at given A
Mutual Information can be regarded as the reduced
uncertainty by the seismic attributes.
Thus, we should choose the one which will maximize the
mutual information.
Rock Physics 9
Case Studies
North Sea Tertiary Turbiditic Reservoir
Case I
- Facies Identification H ( facies | attributes )
Case II
- Pore Fluid Identification H ( fluid | attributes )
References
- Tapan et al. (2001)
- Takahashi et al (1999)
Rock Physics 10
Well Log Data
Facies Classification in Well Log Data
IIa: Cemented Sand
IIb: Uncololidated Sand
IIc: Laminated Sand
III: Interbedded Sand-Shale
IV: Silty Shale
V: Pure Shale
Takahashi et al. (1999) Rock Physics 11
IIa: Cemented Sand
Crossplot IIb: Uncololidated Sand
IIc: Laminated Sand
III: Interbedded Sand-Shale
IV: Silty Shale
V: Pure Shale
Takahashi et al. (1999) Rock Physics 12
Conditional probability distributions
Univariate
Bivariate
Takahashi et al. (1999) Rock Physics 13
Mutual Information
Information about lithofacies carried by Seismic attributes
Takahashi et al. (1999) Rock Physics 14
Case Study 2 (Pore Fluid)
Extended PDFs Gassmann eq. was applied for fluid substitution in only
sand reservoir
Univariate
Bivariate
Vp
Takahashi et al. (1999) Rock Physics 15
Mutual Information
Information about pore fluid carried by Seismic attributes
Takahashi et al. (1999) Rock Physics 16
Discussions
Linear measures of uncertainty, such as variance (σ2) and
covariance (σ12), can be used instead of the entropy (H) ?
Variance (covariance) can work only at limited situation
- Parametric PDFs, such as Gaussian distribution
- Continuous variable
Information Entropy can work more flexibly
- Nonparametric PDFs
- Categorical variables (Shale, Sand)
The Entropy offers a more flexible representation of the
state of information about the rock.
Rock Physics 17
Monte Carlo Simulation
A technique using random
numbers for probabilistic solution
of a model
• Model is nonlinear system
• Input parameters with uncertainty
• Uncertainty analysis instead of
deterministic method
http://www.vertex42.com/ExcelArticles/mc/MonteCarloSimulation.html
Rock Physics 18
Monte Carlo Simulation
Lognormal
Normal Data-deriven nonparametric PDF
Constant etc
Step 1: Create a model, y = f(x1, x2, ..., xq).
Step 2: Generate a set of random inputs, xi1, xi2, ..., xiq.
Step 3: Use the model to obtain outputs.
Step 4: Repeat steps 2 and 3 for i = 1 to n.
Step 5: Analyze the results using histograms, summary
statistics, confidence intervals, etc.
http://www.vertex42.com/ExcelArticles/mc/MonteCarloSimulation.html
Rock Physics 19
Thank you for attentions
Rock Physics 20
