ADAPTIVE LOCAL KRIGING (ALK) TO RETRIEVE THE SLANT RANGE SURFACE MOTION MAPS OF WENCHUAN EARTHQUAKE.pptx
1. ADAPTIVE LOCAL KRIGING (ALK)
TO RETRIEVE THE SLANT RANGE
SURFACE MOTION MAPS OF
WENCHUAN EARTHQUAKE
Department of Earth Science and Engineering
Imperial College London
Meng-Che Wu
meng-che.wu08@imperial.ac.uk
Jian Guo Liu
j.g.liu@imperial.ac.uk
6. Ordinary kriging concept
Ordinary kriging:
Γ*λ=g
Γ is a matrix of the semivariance between each sampled point.
λ is a vector of the kriging weights.
g is a vector of the semivariance between a unknown point and
each sampled point.
Semivariance = FSM(D)
FSM is the fitted semivariogram model.
D is the distance bewteen each sampled point or the distance
between a unknown point and each sampled point.
N
Z(s0 ) Σ λi Z(si ) S = (x, y) is a location
i 1
8. Method: Adaptive Local Kriging
Hang wall 1. Window based
kriging scan to
calculate the linear
fitting of local
semivariance.
2. Window size is
≈1m locally adaptive to
ensure adequate
data points and
high processing
Azimuth
efficiency.
Foot wall ≈ -1 m
Range
9. ALK local semivariogram model:
Towards the seismic fault (Hang
wall side)
Semivariance Local gradient: 1.258 10-5
Distance
Averaged semivariance Fitted semivariance
x = 1024, y = 230
10. ALK local semivariogram model:
Towards the seismic fault (Hang
wall side)
Semivariance Local gradient: 5.812 10-5
Distance
Averaged semivariance Fitted semivariance
x = 1024, y = 460
11. ALK local semivariogram model:
Towards the seismic fault (Hang
wall side)
Semivariance Local gradient: 7.313 10-5
Distance
Averaged semivariance Fitted semivariance
x = 1024, y = 580
12. ALK local semivariogram model:
Towards the seismic fault (Foot
wall side)
Semivariance Local gradient: 1.624 10-5
Distance
Averaged semivariance Fitted semivariance
x = 745, y = 1200
13. ALK local semivariogram model:
Towards the seismic fault (Foot
wall side)
Semivariance Local gradient: 3.613 10-5
Distance
Averaged semivariance Fitted semivariance
x = 745, y = 1000
14. ALK local semivariogram model:
Towards the seismic fault (Foot
wall side)
Semivariance Local gradient: 7.652 10-5
Distance
Averaged semivariance Fitted semivariance
x = 745, y = 870
15. ALK multi- H
Give some sampled
points in the large
step Ordinary decoherence gaps
kriging
processing F Coherence
flow chart thresholding
Input Hang wall Final
Coherence
data & foot wall ALK
thresholding
separation result
ALK
(Decoherence
zone)
H
ALK
F Artificial discontinuity
elimination
19. ALK results assessment
A
Path 471 profiles
A A’
RMSE:
0.0053591572
meters Original unwrapped
Correlation image profile
coefficient:
0.99999985
A’
Azimuth
ALK data profile
Range
≈1m ≈ -1 m
20. ALK results assessment
Path 472 profiles
A A A’
RMSE:
0.00909682429
meters Original unwrapped
Correlation image profile
coefficient:
0.99939712
Azimuth
A’ ALK data profile
Range
≈1m ≈ -1 m
21. ALK results assessment
Path 473 profiles
A A’
A
RMSE:
0.0083477924
meters Original unwrapped
Correlation image profile
coefficient: Traced fault line Initial fault
0.99973365
Azimuth
A’ ALK data profile
Range
≈1m ≈ -1 m
22. ALK results assessment
Path 474 profiles
A A’
A
RMSE:
0.017175553
meters Original unwrapped
Correlation image profile
Traced fault line
coefficient: Initial fault
0.99792644
Azimuth
A’
ALK data profile
Range
≈1m ≈ -1 m
23. ALK results assessment
Path 475 profiles
A A A’
RMSE:
0.0059325138
meters Original unwrapped
Correlation image profile
coefficient: Initial fault Traced fault line
0.99969193
Azimuth
A’
ALK data profile
Range
≈1m ≈ -1 m
24. ALK results assessment
Path 476 profiles
A A’
A
RMSE:
0.0071013203
meters Original unwrapped
Correlation image profile
coefficient:
0.99929831
Azimuth
A’
ALK data profile
Range
≈1m ≈ -1 m
28. 3D view of refined ALK unwrapped data
≈1m
≈ -1 m
29. Conclusions
Local semivariogram is more representive to
the local variation of spatial pattern of the
interferogram than a global semivariogram
model.
Dynamical local linear model represents a
nonlinear global model for the whole
interferogram.
ALK multi-step processing procedure
avoids the error increases in large
decoherence gaps.
30. Conclusions
The ALK interpolation data revealed dense
fringe patterns in the decoherence zone and
show high fidelity to the original data
without obvious smoothing effects.
The initial fault line separating the data does
not affect the final interpolation result of ALK
processing.
The seismic fault line that can be denoted in
the ALK is different from that in publications.
The discrepancy needs further investigation.
31. Future works
Geological structural numerical
modeling to explain the discrepancy
of trend of seismic fault line.
Three dimensional surface
deformation maps development.