Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging...
Transcript of Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging...
![Page 1: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/1.jpg)
113
Best Linear Unbiased Estimation and Kriging
Presented by
Gordon A. Fenton
in Stochastic Analysis and Inverse Modeling
![Page 2: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/2.jpg)
114
Introduction
We often want some way of estimating what is happening at unobserved locations given nearby observations. Common ways of doing so are as follows;
1. Linear Regression: we fit a curve to the data and use that curve to interpolate/extrapolate. Uses only geometry and the data values to determine the curve (by minimizing the sum of squared errors).
2. Best Linear Unbiased Estimation (BLUE): similar to Linear Regression, except that correlation is used, instead of geometry, to obtain the best fit.
3. Kriging: similar to BLUE except the mean is estimated rather than being assumed known.
![Page 3: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/3.jpg)
115
Best Linear Unbiased Estimation (BLUE)
We express our estimate at the unobserved location Xn+1, as a linear combination of our observations, X1, X2, …, Xn; Notice that position does not enter into this estimate at all – we will determine the unknown coefficients, βk , using 1st and 2nd moment information only (i.e. means and covariances). To determine the “Best” estimator, we look at the estimator error, defined as the difference between our estimate and the actual value;
( )1 11
ˆn
n n k k kk
X Xµ β µ+ +=
= + −∑
( )1 1 1 11
ˆn
n n n n k k kk
E X X X Xµ β µ+ + + +=
= − = − − −∑
![Page 4: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/4.jpg)
116
Best Linear Unbiased Estimation (BLUE)
Aside: If the covariances are known, then they include information both about distances between observation points, but also about the effects of differing geological units. Linear regression considers only distance between points. Thus, regression cannot properly account for two observations which are close together, but lie in largely independent layers. The covariance between these two points will be small, so BLUE will properly reflect their effective influence on one another.
![Page 5: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/5.jpg)
117
Best Linear Unbiased Estimation (BLUE)
To make the estimator error as small as possible, its mean should be zero and its variance minimal. The mean is automatically zero by the selected form of the estimator;
( )
[ ]
( )
1 1 1 11
1 11
1
ˆE E
E
0
n
n n n n k k kk
n
n n k k kk
n
k k kk
X X X X
X
µ β µ
µ µ β µ
β µ µ
+ + + +=
+ +=
=
− = − − −
= − − −
= − −
=
∑
∑
∑
In other words, this estimator is unbiased.
![Page 6: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/6.jpg)
118
Best Linear Unbiased Estimation (BLUE)
Now we want to minimize the variance of the estimator error:
( )2
1 1 1 1
2 21 1 1 1
2 21 1 1 1
ˆ ˆVar E
ˆ ˆE 2
ˆ ˆE 2E E
n n n n
n n n n
n n n n
X X X X
X X X X
X X X X
+ + + +
+ + + +
+ + + +
− = − = − +
= − + To simplify the algebra, we will assume that μ = 0 everywhere (this is no loss in generality). In this case, our estimator simplifies to
11
ˆn
n k kk
X Xβ+=
=∑
![Page 7: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/7.jpg)
119
Best Linear Unbiased Estimation (BLUE)
Our estimator error variance becomes
[ ]
[ ]
21 1 1 1
1 1 1
21 1
1 1 1
ˆVar E 2 E E
Var 2 Cov , Cov ,
n n n
n n n k n k k j k jk k j
n n n
n k n k k j k jk k j
X X X X X X X
X X X X X
β β β
β β β
+ + + += = =
+ += = =
− = − +
= − +
∑ ∑∑
∑ ∑∑
We minimize this with respect to our unknown coefficients, β1, β2, …, βn, by setting derivatives to zero;
1 1ˆVar 0 for 1,2, ,n n
l
X X l nβ + +
∂ − = = ∂
![Page 8: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/8.jpg)
120
Best Linear Unbiased Estimation (BLUE)
[ ]
[ ] [ ]
[ ]
1
1 11
1 1 1 1
Var 0
Cov , Cov ,
Cov , 2 Cov , 2
nl
n
k n k n l lkl
n n n n
j k j k k l k k lkk j k kl
X
X X X X b
X X X X C
β
ββ
β β β ββ
+
+ +=
= = = =
∂=
∂
∂= =
∂
∂ = = ∂
∑
∑∑ ∑ ∑
so that 1 11
ˆVar 2 2 0n
n n l k lkkl
X X b Cββ + +
=
∂ − = − + = ∂ ∑
in other words, is the solution to 1 β β −= → =C b C b
β
![Page 9: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/9.jpg)
121
BLUE: Example
Suppose that ground penetrating radar suggests that the mean depth to bedrock, μ, in metres, shows a slow increase with distance, s, in metres, along the proposed line of a roadway.
![Page 10: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/10.jpg)
122
BLUE: Example
However, because of various impediments, we are unable to measure beyond s = 20 m. The best estimate of the depth to bedrock at s = 30 m is desired.
![Page 11: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/11.jpg)
123
BLUE: Example
Suppose that the following covariance function has been established regarding the variation of bedrock depth with distance where σX = 5 m and τ is the separation distance between points. We want to estimate the bedrock depth, X3, at s = 30 m, given the following observations of X1 and X2 at s = 10 m and 20 m, respectively : at s = 10 m, x1 = 21.3 m at s = 20 m, x2 = 23.2 m
( ) 2 exp40XCτ
τ σ
= −
![Page 12: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/12.jpg)
124
BLUE: Example
Solution: We start by finding the components of the covariance matrix and vector:
[ ][ ]
20/ 401 3 2
10/ 402 3
Cov ,Cov , X
X X eX X e
σ−
−
= =
b
[ ] [ ][ ] [ ]
10/ 401 1 1 2 2
10/ 402 1 2 2
Cov , Cov , 1Cov , Cov , 1X
X X X X eX X X X e
σ−
−
= =
C
Note that b contains the covariances between the observation and prediction points, while C contains the covariances between observation points only. This means that it is simple to compute predictions at other points (C only needs to be inverted once).
![Page 13: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/13.jpg)
125
BLUE: Example
Now we want to solve for the unknown coefficients Notice that the variance cancels out – this is typical of stationary processes. Solving, gives us
10/ 40 20/ 4012 2
10/ 40 10/ 402
11X X
e ee e
βσ σ
β
− −
− −
=
110/ 40 20/ 401
10/ 4010/ 40 10/ 402
011
e eee e
ββ
−− −
−− −
= =
Thus, β1 = 0 and β2 = e−10/40. Note the Markov property: the ‘future’ (X3) depends only on the most recent past (X2) and not on the more distance past (X1).
β =C b β
![Page 14: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/14.jpg)
126
BLUE: Example
The optimal linear estimate of X3 is thus
( ) ( )( )( ) ( )( )
10/ 403 2
1/ 4
1/ 4
ˆ 30 20
20 0.3 30 23.2 20 0.3 20
29.0 2.826.8 m
x e x
e
e
µ µ−
−
−
= + −
= + + − − = −=
![Page 15: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/15.jpg)
127
BLUE: Estimator Error
Once the best linear unbiased estimate has been determined, it is of interest to ask how confident are we in our estimate? If we reconsider our zero mean process, then our estimator is given by which has variance
11
ˆn
n k kk
X Xβ+=
=∑
2ˆ1
1
1 1
ˆVar =Var
Cov ,
n
n k kXk
n n
k j k jk j
T T
X X
X X
σ β
β β
β β β
+=
= =
=
=
= =
∑
∑∑C b
![Page 16: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/16.jpg)
128
BLUE: Estimator Error
The estimator variance is often more of academic interest. We are typically more interested in asking questions such as: What is the probability that the true value of Xn+1 exceeds our estimate by a certain amount? For example, we may want to compute where b is some constant. To determine this, we need to know the distribution of the estimator error If X is normally distributed, then our estimator error is also normally distributed with mean zero, since unbiased,
1 1 1 1ˆ ˆP Pn n n nX X b X X b+ + + +
> + = − >
1 1ˆ( )n nX X+ +−
1 1ˆ = E 0E n nX Xµ + +
− =
![Page 17: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/17.jpg)
129
BLUE: Estimator Error
The variance of the estimator error is
[ ]2 21 1
1 1 1
2
2
2
2
Var 2 Cov , Cov ,
2
( )
n n n
E n k n k k j k jk k j
T TX
T T TX
T TX
TX
X X X X Xσ β β β
σ β β β
σ β β β β
σ β β β
σ β
+ += = =
= − +
= + −
= + − −
= + − −
= −
∑ ∑∑C bC b bC b b
b
where we made use of the fact that β is the solution to Cβ = b, or, equivalently, Cβ − b = 0
![Page 18: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/18.jpg)
130
BLUE: Conditional Distribution
The estimator is also the conditional mean of Xn+1 given the observations. That is,
1ˆ
nX +
[ ]1 1 2 1ˆE | , , ,n n nX X X X X+ +=
The condition variance of Xn+1 is just the variance of the estimator error;
[ ] 21 1 2Var | , , ,n n EX X X X σ+ =
In general, questions regarding the probability that Xn+1 lies in some region should employ the conditional mean and variance of Xn+1, since this makes use of all of the information.
![Page 19: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/19.jpg)
131
BLUE: Example
Consider again the previous example. 1. compute the variance of the estimator and the estimator error 2. estimate the probability that X3 exceeds by more than 4 m. 3X̂
![Page 20: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/20.jpg)
132
BLUE: Example
Solution: We had
1/42
1/4
11X
ee
σ−
−
=
C
and 2/4
210/40 1/4
0, X
ee e
β σ−
− −
= =
b
so that { }2/4
2 2 1/4 2 2/4ˆ 3 1/4
ˆVar 5 0 5TX
eX e e
eσ β
−− −
−
= = = = b
which gives 1/ 4ˆ 5 3.894 mX eσ −= =
![Page 21: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/21.jpg)
133
BLUE: Example
The covariance vector found previously was 2/ 4
21/ 4X
ee
σ−
−
=
b
The variance of the estimator error is then
{ }
( )
2 23 3
2/ 42 2 1/ 4
1/ 4
2 2/ 4
ˆVar
0
5 1
TE X
X X
X X
ee
e
e
σ σ β
σ σ−
−−
−
= − = −
= −
= −
b
The standard deviation is thus 2/ 45 1 3.136 mE eσ −= − =
This is less than the estimator variability and significantly less than the variability of X (σX = 5). This is due to the restraining effect of correlation between points.
![Page 22: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/22.jpg)
134
BLUE: Example
We wish to compute the probability We first need to assume a distribution for Let us assume that X is normally distributed. Then since the estimator is simply a sum of X’s, it too must be normally distributed. This, in turn implies that the quantity is normally distributed. Since is an unbiased estimate of X3, and we have just computed σE = 3.136 m. Thus,
3 3ˆP 4X X − >
( )3 3ˆX X−
( )3 3ˆX X−
3X̂
3X̂ 3 3ˆE 0E X Xµ = − =
3 34 4 0ˆP 4 P 1 0.1003
3.136E
E
X X Z µσ
− − − > = > = −Φ =
![Page 23: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/23.jpg)
135
Geostatistics: Kriging
Kriging (named after Danie Krige, South Africa, 1951) is basically BLUE with the added ability to estimate the mean. The purpose is to provide a best estimate of the random field at unobserved points. The kriged estimate is, again, a linear combination of the observations, where x is the spatial position of the unobserved value being estimated. The unknown coefficients, β, are determined by considering the covariance between the observations and the prediction point.
1
ˆ ( )n
k kk
X Xβ=
=∑x
![Page 24: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/24.jpg)
136
Geostatistics: Kriging
In Kriging, the mean is expressed as a regression where gi(x) is a specified function of spatial position x. Usually g1(x) = 1, g2(x) = x, g3(x) = x2, and so on in 1-D. Similarly in higher dimensions. As in a regression analysis, the gi(x) functions should be (largely) linearly independent over the domain of the regression (i.e. the site). The unknown Kriging weights, β, are obtained as the solution to the matrix equation
1( ) ( )
m
X i ii
a gµ=
=∑x x
β =K M
![Page 25: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/25.jpg)
137
Geostatistics: Kriging
where K and M depend on the mean and covariance structure. In detail, K has the form
where Cij is the covariance between Xi and Xj.
![Page 26: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/26.jpg)
138
Geostatistics: Kriging
The vectors β and M have the form
where ηi are Lagrangian parameters used to solve the variance minimization problem subject to the non-bias conditions and Cix are the covariances between the ith observation point and the prediction location, x.
![Page 27: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/27.jpg)
139
Geostatistics: Kriging
The matrix K is purely a function of the observation point locations and their covariances – it can be inverted once and then used repeatedly to produce a field of best estimates (at each prediction point, only the RHS vector M changes). The Kriging method depends on
1. knowledge of how the mean varies functionally with position (i.e., g1, g2, … need to be specified), and
2. knowledge of the covariance structure of the field. Usually, assuming a mean which is constant (m = 1, g1(x) = 1, a1 = μX ), or linearly varying is sufficient.
![Page 28: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/28.jpg)
140
Geostatistics: Kriging
Estimator Error: The estimator error is the difference between the true (random) value X(x) and its estimate . The estimator is unbiased, so that and its variance is given by where βn and Mn are the first n elements of the vectors β and M, and Kn×n is the n × n upper left submatrix of K containing the covariances. As with BLUE, is the conditional mean of X(x). The conditional variance of X(x) is .
ˆ ( )X x
( ) ( )ˆE 0E X Xµ = − = x x
( ) ( )( ) ( )22 2ˆE 2T
E X n n n n nX Xσ σ β β× = − = + −
x x K M
ˆ ( )X x2Eσ
![Page 29: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/29.jpg)
141
Example: Foundation Consolidation Settlement
Consider the estimation of consolidation settlement under a footing. Assume that soil samples/tests have been obtained at 4 nearby locations.
![Page 30: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/30.jpg)
142
Example: Foundation Consolidation Settlement
The samples and local stratigraphy are used to estimate the soil parameters Cc, eo, H, and po appearing in the consolidation settlement equation where S = settlement N = model error random variable (μN = 1.0, σN = 0.1) eo = initial void ratio Cc = compression index po = initial effective overburden stress Δp = mid-depth stress increase due to applied footing load H = depth to bedrock
10log1
c o
o o
C p pS N He p
+ ∆= +
![Page 31: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/31.jpg)
143
Example: Foundation Consolidation Settlement
We will assume that the estimation error in obtaining the soil parameters from the samples is negligible compared to field variability, so this source of error will be ignored. We do, however, include model error through a multiplicative random variable, N, which is assumed to have mean 1.0 (i.e. the model correctly predicts the settlement on average) and standard deviation 0.1 (i.e. the model has a standard error of 10%). The mid-depth stress increase due to the footing load, Δp, is assumed to be random with
[ ]p
E 25 kPa 5 kPa
pσ∆
∆ =
=
![Page 32: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/32.jpg)
144
Example: Foundation Consolidation Settlement
Assume that all four random fields (Cc, eo, H, and po) are stationary and that the correlation function is estimated from similar sites to be
( ) 2, exp
60i j
i jρ − = −
x xx x
![Page 33: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/33.jpg)
145
Example: Foundation Consolidation Settlement
Using the correlation function, we get the following correlation matrix between sample locations:
We will assume that the same correlation length (θ = 60 m) applies to all 4 soil parameters. Thus, for example, the covariance matrix between sample points for Cc is [ ]2
cCσ ρ
We will obtain Kriging estimates from each of the four random fields independently – if cross-correlations between the parameters are known, then the method of co-Kriging can be applied (this is essentially the same, except with a much larger cross-covariance matrix).
![Page 34: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/34.jpg)
146
Example: Foundation Consolidation Settlement
The Kriging matrix associated with the depth to bedrock, H, is
where we assumed stationarity, with m = 1 and g1(x) = 1. If we place the coordinate axis origin at sample location 4, the footing has coordinates x = (20, 15) m. The RHS vector for H is
![Page 35: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/35.jpg)
147
Example: Foundation Consolidation Settlement
Solving the matrix equation gives the following four weights;
H H Hβ =K M
We can see from this that samples closest (most highly correlated) to the footing are weighted the most heavily. (i.e. sample 4 is closest to the footing) All four soil parameters will have identical weights.
![Page 36: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/36.jpg)
148
Example: Foundation Consolidation Settlement
The best estimates of the soil parameters at the footing location are thus,
The estimation errors are given by
( )2 2 2TE X n n n n nσ σ β β×= + −K M
Since Kn×n is just the correlation matrix, ρ, times the appropriate soil parameter variance (which replaces ), and similarly Mn is just the correlation vector times the appropriate variance, the variance can be factored out
2Xσ
( )2 2 1 2TE X n nσ σ β ρβ ρ = + − x
![Page 37: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/37.jpg)
149
Example: Foundation Consolidation Settlement
( )2 2 1 2TE X n nσ σ β ρβ ρ = + − x
where ρx is the vector of correlation coefficients between the samples and the footing. For the Kriging weights and given correlation function, this gives
2 20.719E Xσ σ=
The individual parameter estimation errors are thus
![Page 38: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/38.jpg)
150
Example: Foundation Consolidation Settlement
In summary, the variables entering the consolidation settlement formula have the following statistics based on the Kriging analysis:
where v is the coefficient of variation (σ/μ).
![Page 39: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/39.jpg)
151
Example: Foundation Consolidation Settlement
A first-order approximation to the mean settlement is obtained by substituting the mean parameters into the settlement equation;
100.386 177.3 25(1.0) (4.30) log 0.0434 m
1 1.19 177.3Sµ+ = = +
![Page 40: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/40.jpg)
152
Example: Foundation Consolidation Settlement
A first-order approximation to the variance of settlement is given by
26
2
1jS X
j j
SX
µ
σ σ=
∂= ∂ ∑
where Xj is replaced by the 6 random variables, N, Cc, etc, in turn. The subscript μ means that the derivative is evaluated at the mean of all random variables.
![Page 41: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/41.jpg)
153
Example: Foundation Consolidation Settlement
Evaluation of the partial derivatives of S with respect to each random variable gives
so that 26
2 5 2
118.952 10 m
jS Xj j
SX
µ
σ σ −
=
∂= = × ∂ ∑
0.0138 mSσ =
![Page 42: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/42.jpg)
154
Example: Foundation Consolidation Settlement
We can use these results to estimate the probability of settlement failure. If the maximum settlement is 0.075 m, and we assume that settlement, S, is normally distributed with mean 0.0434 m and standard deviation 0.0138, then the probability of settlement failure is
[ ] ( )0.075 0.0434P 0.075 1 1 2.29 0.010.0138
S − > = −Φ = −Φ =
![Page 43: Best Linear Unbiased Estimation and Kriging in …113 Best Linear Unbiased Estimation and Kriging Presented by . Gordon A. Fenton . in . Stochastic Analysis and Inverse Modelingalertgeomaterials.eu/data/school/2014/session2/06b_Fenton_Blue.pdf ·](https://reader034.fdocuments.in/reader034/viewer/2022052313/5aaf5aa67f8b9a59478d35d2/html5/thumbnails/43.jpg)
155
Conclusions
• Best Linear Unbiased Estimation requires prior knowledge of the mean and the covariance structure of the random field(s)
• if the mean and covariance structure are known, then BLUE is superior to simple linear regression since it more accurately reflects the dependence between random variables than does simply distance
• Kriging is a variant of BLUE in which the mean is estimated from the observations as part of the unbiased estimation process.