You are on page 1of 5

MIN E 613 - N ON -PARAMETRIC AND M ULTIVARIATE G EOSTATISTICS

Assignment 1

Thiago Alduini Mizuno


September 15, 2021

1 B ASIC NOTATION OF G EOSTATISTICS


The section presents a list of the main notation used in Geostatistics. The list is based on
the book GSLIB: Geostatistical Software Library and User’s Guide from the authors Clayton V.
Deutsch and Andre G. Journel, 1997. Original LaTeX files from CCG knowledge base were used
to create the notations.

1.0.1 R ANDOM VARIABLE


Z random variable (RV)
z a value of Z, an outcome
(u) location, coordinate
Z(u) RV in the location (u)
z(u) a value of Z in the location (u)
E{Z} Expected value of Z

F (u; z) = Prob {Z (u) ≤ z} (1.1)

Cumulative distribution function (CDF) of Z(u)

F (u; z|(n)) = Prob {Z (u) ≤ z|(n)} (1.2)

CDF of Z(u) for a particular set data where (n) are neighboring data values

F (u; k|(n)) = Prob {Z (u) = k|(n)} (1.3)

CDF of a Z(u) for a categorical data k for the categories.

1
1.0.2 R ANDOM F UNCTION

RF Set of random variables Z(u) defined in a field of interest

F (u1 , . . . , uk ; z 1 , . . . , z K ) = Prob {Z (u1 ) ≤ z 1 , . . . , Z (uK ) ≤ z K } (1.4)

CDF of a random function

F (u, u0 ; z, z 0 ) = Prob Z (u) ≤ z, Z (u0 ) ≤ z 0


© ª
(1.5)

Particular case of CDF for bivariate RF


½
1, if Z (u) ≤ z
I (u; z) = (1.6)
0, otherwise

Binary transformation of Z(u) in the Indicator I(u;z)

F (u, u0 ; z, z 0 ) = E I (u; z)I (u0 ; z 0 )


© ª
(1.7)

Indicator covariance bivariate case

C (u, u0 ) = E Z (u)Z (u0 ) − E {Z (u)} E Z (u0 )


© ª © ª
(1.8)

Covariance between Z(u) and Z(u’)

1.0.3 VARIOGRAM
γ(h) variogram
(h) lag distance
C(0) stationary variance
C(h) stationary covariance

2γ(h) = Var {Z (u + h) − Z (u)} (1.9)


γ(h) = C (0) −C (h), ∀ u

Variogram for stationary RF

1.0.4 K RIGING
Z∗SK Simple kriging estimator
λ weigths
m mean

n µ n ¶

λα (u)Z (uα ) + 1 − λα (u) m
X X
ZSK (u) = (1.10)
α=1 α=1

Simple kriging equation

2
n
λβ (u)C (uβ − uα ) = C (u − uα ),
X
(1.11)
β=1

Set of normal equations that minimize the error variance


n
σ2SK (u) = C (0) − λα (u)C (u − uα ) ≥ 0
X
(1.12)
α=1

Kriging variance

E {I (u; z)|(n)} = 1 · Prob {I (u; z) = 1|(n)}


+0 · Prob {I (u; z) = 0|(n)}
= 1 · Prob {Z (u) ≤ z)|(n)} ≡ F (u; z|(n)), as defined in (II.2)

Indicator Kriging Approach

1.0.5 S TOCHASTIC S IMULATION


(l )
{z (u), u ∈ A}, l = 1, . . . , L realizations of a stochastic simulation
L one possible set of z(u) on a field A

2 C ORRELATION MATRIX
This section analyses the possible correlation matrices for three variables to investigate if the
matrices are valid: positive and definite. Matrix sets were created using a 0.05 interval between
the correlation values of the three variables. For each matrix, the eigenvalue was calculated to
define its validity.

3
Figure 2.1: Jupyter code to validate the matrices and 3D plot of valid points

3 M ODIFICATION OF B LOCK SIS PROGRAM (GSLIB)


The original code of Block SIS from GSLIB requires two variograms to run a simulation with
two categories. However, in the presence of two categories, it is necessary only one variogram
perform the simulation. This section will describe the modification of the Fortran code to edit
this option. The version used was the 1.5.0 — Clayton V. Deutsch, latest modification from
Ryan M. Barnett, May 30, 2018, Source and Executable: blocksis 150.zip. From CCG knowledge
base.
Parameters are loaded via a file. The number of categories is defined in line 164 as variable
ncat. The variogram parameters are load from line 333 to 357, the modification on the code
exits the loop if the category number ( variable ncat) equals 2. After that, the code defines the
same parameters for the variogram of both categories.

4
Figure 3.1: Original code for variogram load

Figure 3.2: Edited code and program after modification

You might also like