Scribd
Upload
63 views

Geostatistics Formula Sheet

Formulas Geoestadisticas

Uploaded by

Anonymous cMBxGA
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
63 views

Geostatistics Formula Sheet

Formulas Geoestadisticas

Uploaded by

Anonymous cMBxGA
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

GeostatisticsFormulaSheet

June2008

Z denotes a random variable (RV). z denotes an outcome.


Z(u) denotes a regionalized RV at location u. The set of
random variables over a stationary domain A {Z (u ),uA} is
knownasarandomfunction(RF).
Uncertainty in a RV is represented by a cumulative
distributionfunction(CDF):F(z)=Prob{Zz}.Thederivativeof
the CDF is the probability density function (PDF): f(z)=F(z).
Quantiles are zvalues with a probabilistic meaning: zp such
thatF(zp)=p.ThequantilefunctionisdenotedF1(p)=zp.

( y) =
F
Y |Z1,, Z N

The covariance and correlation coefficient summarize


bivariatedependencebetweentworandomvariables:
Cov{ X , Y } = C

mean. The variance is 2 = E{[ Z m]2 }= E{Z 2 } m2 . is the


standarddeviation./misthecoefficientofvariation.

XY = = C XY / ( X iY )

Zisuniformintheintervalatobwhen:

2
2 (h ) = E [ Z (u ) Z (u +h )]

ZisstandardnormalorGaussianwhen f ( z ) =

1
e
2

Understationaritythevariogram,varianceandcovarianceare
relatedby(h)=2C(h).
2

h h h
Thescalarnormalizeddistanceis h = X + Y + Z
aX

aY

aZ

RotationofX/Ybyangleisachievedby:
x1 cos
y = sin
1

2
1/(b a ), z[ a ,b ]
a +b
2 (b a )

, m=
and =
f (z) =
2
12
0, otherwise

ThevariogramforalaghIsdefinedas

ArandomvariableZisstandardizedbyY=(ZmZ)/.E{Y}=0,
E{Y2}=1andZ=Y +mZ.

z2
2

= E{[ X m ][Y m ]} = E{ XY } m i m
X
Y
X Y

XY

z i f ( z ) dz .
Theexpectedvalueoperatoriswritten E{Z } =

E{Z} is denoted m and is also known as the first moment or

FY , Z1,, Z N ( y , z1,, z N )

FZ1,, Z N ( z1,, z N )

sin x0

cos y0

Stratigraphic coordinate is calculated with correlation base,


topandaveragethicknessas:

Z rel ( x, y )=

Z ( x , y ) Zcb ( x , y )
iT
Zct ( x , y ) Zcb ( x , y )

Variogramsaremodeledbystructures: (h ) = inst
= 0 Ci i i (h ) .
Common standardized models include the Exponential
Exp ( h ) =1exp( 3h / a ) ,Spherical Sph ( h ) = 1.5( h / a ) 0.5( h / a )

2
if h a; 1, otherwise , Gaussian Gaus ( h ) = 1 exp( 3( h / a ) ) .

Theholeeffectislesscommon: ( h )=C i 1cos


a

ThevolumeaveragedvariogrambetweenvandV(gammabar):

ThevariableZ>0islognormalwithmand 2whenY=ln(Z)is
normalwithmeanandvariance2.Theparameters:
2

= ln( m ) / 2
+ 2 /2

m=e

= ln 1+ 2 / m2

2 2
= m e 1

(V ,v )=

Thedispersionvarianceisgivenby:

( v,V ) = E [ Zv mV ]2
2

} = (V ,V ) ( v,v )

Variancesadd: D ( v , A ) = D ( v ,V ) + D (V , A )

ThemultivariatedistributionofNRVsZi,i=1,...,Nisdefinedas:

v <V < A

Varianceofalinearcombination:

F
( z , , z ) = Prob {Z1 z1,, Z N z N }
Z1,, Z N 1
N

x =

ConditionaldistributionsarecalculatedwithBayesLaw:

1
( x y )dxdy
|V |i|v|V v

1 n
x
n i =1 i

Var{ x } =

n
n
X + 1 Cov{ x , x }
i j
n
n 2 i =1 j =1
i j

Page1of2

ThenvariatemultivariateGaussiandistributionisdefined:

Linearestimationatu.givenby: z* m = i i[ zi mi ]
i =1

f (y ) =

Theestimationvarianceiscalculatedas:
n

n n

i =1

i =1 j =1

Minimizing the estimation variance leads to simple kriging


andminimizedestimationvariance(krigingvariance):
i =1,...,n

2 = 2 C
SK
i i,

LUsimulationfromacovariancematrix:C=LU;y=Lw.
Sequential simulation relies on recursive decomposition of
themultivariatedistributionwithBayeslaw:
P ( A , ..., A ) = P ( A | A , ..., A
) P ( A , ..., A
)
N
N
N 1
N 1
1
1
1

i =1

= P ( AN | A1 ,..., AN 1 ) P ( AN 1| A1 ,..., AN 2 ) P ( A1 ,..., AN 2 )

Ordinarykrigingconstrainthesumoftheweightstoone:
n
j Ci , j + =Ci ,
j =1

n
j =1
j =1

Where is the 1xn vector of mean values and is the nxn


matrix of covariances. Conditional distributions defined by
normalequations(seesimplekriging).

2 = 2 2 C +
E
i i , i j Ci , j

n
j Ci , j =Ci ,
j =1

1
1

T
exp ( y ) 1 ( y )
n 1/2
2

2 ||

i =1,...,n

= P ( AN | A1 ,..., AN 1 ) P ( AN 1| A1 ,..., AN 2 ) P ( A2 | A1 ) P ( A1 )

Simulationfromaunivariatedistributionamountstoquantile
transformationofarandomnumber:
1
z = F (r)
s
Z

Universalkriging/krigingwithatrendconstrainthemeanto
L

Indicatorsforcontinuousvariables

theform m (u ) = al i fl (u )
l =0

n
L
j Ci , j + l =Ci ,
j =1
l =0

j i fl ( u j ) = fl ( u )
j =1

1, if z (u ) zc
i ( u ; zc ) =
0, otherwise

i =1,...,n

1, if u k
i ( u ;k ) =
0, otherwise

l =0,...., L

Matrixofcrossvariogramscanbemodeledbylinearmodelof
coregionalization(LMC)i,j=1,...,M:

Stepwiseconditionaltransformation:

K
2 (h ) = bik, j i k ( h )
i, j
k =0

Y = G
1

Where each MxM matrix of coefficients (k=0,...,K) must be


positivedefinite.Intrinsicmodelassumesallvariogramsare
proportional. The Markov models assume that the cross
variogram/covarianceisproportionaltoadirectvariogram.

= G
Y
3|2,1

( F ( z )) and z = F

Prob( Z3 z3|Y2 = y2 ,Y1= y1 )

BayesianupdatingpriorandlikelihoodGaussiandistributions:
yU =

Cokrigingconsiderscorrectcovariancebetweendataevents.
ZdataaretransformedtobeYnormal(G(y)isGaussianCDF)
withnormalscoretransform:

Prob( Z1 z1 )

Y2|1 = G 1 Prob( Z 2 z2 |Y1= y1 )

C (h ) = b iC (h ) where b = i / j i
i, j
i ,i
i, j

1 P ( A)
P ( A)
P ( A | B , i = 1, ..., N ) =

i
1 P ( A) n 1 P ( A|Bi )

P ( A) i =1 P ( A|Bi )

(h ) = E [ Zi (u ) Zi (u +h )]i[ Z j (u ) Z j (u +h )]

y=G

Var i( u ;k ) = p (1 pk )
k

Permanenceofratiosforcombiningconditionalprobabilities:

ThecrossvariogrambetweenvariableZi(u)andZj(u):

E i( u ;k ) = p
k

{ }

for k = 1, ..., K

Meanandvarianceofanindicatorvariablearegivenby:

2
2
VarianceofSKestimates: Var z*SK = SK .

i, j

for many cutoffs z


c

Indicatorsforcategoricalvariables

Externaldriftconsiders m ( u ) = a0 + a1 f1 (u )

2 +y 2
yL P
P L
2
2
2
P P L + L2

U =

2 2
P
L

2 2 2 + 2
P
P L L

Disclaimer:theremaybemistakesonthisformulasheet.Anymistakesare
yourfaultyoushouldnotneedaformulasheetanyway.

(G ( y ))

Copyright2008ClaytonV.Deutsch

Page2of2

You might also like