spatial interpolation, kriging
TRANSCRIPT
-
7/28/2019 Spatial Interpolation, Kriging
1/30
Spatial Interpolation,
Kriging
Thomas K. Windholz
-
7/28/2019 Spatial Interpolation, Kriging
2/30
Outline
Spatial Interpolation Basics
Simple, Ordinary, and Universal Kriging
Flavors of Kriging
-
7/28/2019 Spatial Interpolation, Kriging
3/30
Spatial Interpolation Basics
Spatial interpolation allows us predictsvalues at unsampled locations.
In general, a model fitting samples canbe split into first and second ordercomponents.
First order can be captured by, forexample, a trend surface throughregression.
Second order looks at residuals and their
Covariance (e.g., Krigingnamed after
-
7/28/2019 Spatial Interpolation, Kriging
4/30
Outline
Spatial Interpolation Basics
Simple, Ordinary, and Universal Kriging
Flavors of Kriging
-
7/28/2019 Spatial Interpolation, Kriging
5/30
Simple, Ordinary, and
Universal Kriging
What is the difference?
Derivation of simple kriging
Equations for prediction & kriging
variance
Analysis steps
Augmentation to incorporate trend
-
7/28/2019 Spatial Interpolation, Kriging
6/30
Differences Among Kriging
Methods Remember equation for a spatial random
variable:
Simple Kriging: No trend (only )
Ordinary Kriging: Constant trend
Universal Kriging: Polynomial trend
)()()( sUsxsYT
)(sU
-
7/28/2019 Spatial Interpolation, Kriging
7/30
Derivation of Simple Kriging
Basic Idea:
n
iii sUssU 1 )()()(
s1
ss2
s3
s4
1
3 4
2
-
7/28/2019 Spatial Interpolation, Kriging
8/30
Derivation of Simple Kriging
Basic Idea:
How should differ from ?)(
sU
s1
ss2
s3
s4
)(sU
1
3 4
2
-
7/28/2019 Spatial Interpolation, Kriging
9/30
Derivation of Simple Kriging
One descriptor is expected mean square
error:
)(
)(2)()(
)()( 22
2
sUsUEsUEsUEsUsUE
n
i
ii
n
i
n
j
jiji ssCsssCss1
2
1 1
),()(2),()()(
)()(2)()( 2 scssCs TT
-
7/28/2019 Spatial Interpolation, Kriging
10/30
Derivation of Simple Kriging
One descriptor is expected mean square
error:
)(
)(2)()(
)()( 22
2
sUsUEsUEsUEsUsUE
)()(2)()( 2 scssCs TT
minimize
n
i
ii
n
i
n
j
jiji ssCsssCss1
2
1 1
),()(2),()()(
-
7/28/2019 Spatial Interpolation, Kriging
11/30
Derivation of Simple Kriging
Minimization (through differentiation) results
in:
which we can use back in our starting
equation of:
)()( 1 scCs
n
i
ii sUssU1
)()()(
-
7/28/2019 Spatial Interpolation, Kriging
12/30
Derivation of Simple Kriging
Note on the side:
),( ji ssC
),( issc
Covariance matrix among all sample sites
Covariance vector between
prediction locations and all sample sitessi
-
7/28/2019 Spatial Interpolation, Kriging
13/30
Kriging variance
The kriging variance results in
(substitute (s) in the expected mean
square error):
)()()()( 1222 scCscsUsUE Te
-
7/28/2019 Spatial Interpolation, Kriging
14/30
Analysis Steps
Remove trend if it exists
Calculate empirical variogram on
residuals
Fit theoretical variogram
Calculate C and c (actually -1 and )
Predict value & add trend
Estimate error
-
7/28/2019 Spatial Interpolation, Kriging
15/30
Augmentation to incorporate
trend
Ordinary kriging incorporates a constant
trend Universal kriging incorporates a trend of
order x
-
7/28/2019 Spatial Interpolation, Kriging
16/30
Ordinary Kriging
General:
Simple Kriging:
Ordinary & Universal Kriging:
)()()( sUsxsY T
n
i
ii sUssU1
)()()(
n
i
ii sYssY1
)()()(
-
7/28/2019 Spatial Interpolation, Kriging
17/30
Ordinary Kriging
Can handle constant trend (mean)
through an augmented matrix C+ and
augmented vectors +(s) and c+(s).
C )(s =
=
)(sc
011
1),(),(
1),(),(
1
111
nnn
n
ssCssC
ssCssC
)(
)(
)(1
s
s
s
n
1
),(
),( 1
nssC
ssC
-
7/28/2019 Spatial Interpolation, Kriging
18/30
Ordinary Kriging
Can handle constant trend (mean)
through an augmented matrix C+ and
augmented vectors +(s) and c+(s).
C )(s =
=
)(sc
011
1),(),(
1),(),(
1
111
nnn
n
ssCssC
ssCssC
)(
)(
)(1
s
s
s
n
1
),(
),( 1
nssC
ssC
Simple kriging
-
7/28/2019 Spatial Interpolation, Kriging
19/30
Ordinary Kriging
Constant trend will be simultaneouslypredicted.
Can estimate variogram from yvalueswithout removing trend (since it isconstant).
Usually works within a neighborhood andnot with entire dataset.
Since it works in a neighborhood trend onlyhas to be constant in the neighborhood (be
cautious with this statement)
-
7/28/2019 Spatial Interpolation, Kriging
20/30
Universal Kriging
Can handle polynomial trend:
C )(s = )(sc
00)()(
00)()(
)()(),(),(
)()(),(),(
1
111
11
111111
npp
n
npnnnn
pn
sxsx
sxsx
sxsxssCssC
sxsxssCssC
)(
)(
)(
)(
1
1
s
s
s
s
p
n
=
)(
)(
),(
),(
1
1
sx
sx
ssC
ssC
p
n
-
7/28/2019 Spatial Interpolation, Kriging
21/30
Universal Kriging
Polynomial trend will be simultaneouslypredicted.
Cannot estimate variogram from yvalues without removing trend first!!!
Thus, trend has to be removed firstanyway to estimate variogram!
Neighborhood not such an issue as withordinary kriging.
-
7/28/2019 Spatial Interpolation, Kriging
22/30
Outline
Spatial Interpolation Basics
Simple, Ordinary, and Universal Kriging
Flavors of Kriging
-
7/28/2019 Spatial Interpolation, Kriging
23/30
Flavors of Kriging
Block Kriging
Co-Kriging
Others: Robust, Disjunctive, and
Indicator Kriging
-
7/28/2019 Spatial Interpolation, Kriging
24/30
Block Kriging
Uses a blockA rather than locations s.
For example:
),( issCA
dsssC
sAC Ai
i
),(
),(
),( ssC2
),(),(
A
sdsdssCAAC A
-
7/28/2019 Spatial Interpolation, Kriging
25/30
Co-Kriging
Basic idea is to use a second, highly
correlated, variable in locations where
primary variable is (or cannot) bemeasured.
Primary and secondary variable
Samples with:
-
7/28/2019 Spatial Interpolation, Kriging
26/30
Co-Kriging
Basic idea is to use a second, highly
correlated, variable in locations where
primary variable is (or cannot) bemeasured.
Primary and secondary variable
Secondary variable only
Samples with:
-
7/28/2019 Spatial Interpolation, Kriging
27/30
Co-Kriging
Essential component of co-kriging is the
cross-covariogram or cross-variogram:
XYYX sXhsYEhC )()()(
)()()()()(2 sXsXsYhsYEhYX
-
7/28/2019 Spatial Interpolation, Kriging
28/30
Co-Kriging
and the empirical cross-variogram
can be estimated through:
hss
jijiYX
ji
xxyyhn
h ))(()(
1)(2
-
7/28/2019 Spatial Interpolation, Kriging
29/30
Co-Kriging
Thus, we can model the variable Y(s)
through:
With the solution foragain as an
augmented system +.
mn
j
jxi
n
i
iyi sYssYssY11
)()()()()(
-
7/28/2019 Spatial Interpolation, Kriging
30/30
Spatial Interpolation,
Kriging
Thomas K. Windholz