1. 74th EAGE Conference & Exhibition
incorporating SPE EUROPEC 2012
Automated seismic-to-well ties?
Roberto H. Herrera and Mirko van der Baan
University of Alberta, Edmonton, Canada
rhherrer@ualberta.ca
2. Outline
• Introduction
• Similarity in time series
• The manual seismic-to-well tie
• What is Dynamic Time Warping
– How DTW works?
– The automated approach
• Real examples
– Manual vs Automatic
• Conclusions
3. Seismic-to-well similarity
• Objective:
– Can you automate the seismic-to-well tie?
• Possible applications:
– Seismic-to-well tie, log-to-log correlation,
alignment of baseline + monitor in 4D
• Main problem
– Bulk shift, stretching and squeezing is an
interpretation item.
–How to implement semi-automatically?
8. Common similarity measures
Cross-correlation
1
2 2 1/2
1 1
[ ( ) ][ ( ) ]
( )
( [ ( ) ] [ ( ) ] )
n
S T
i
ST n n
S T
i i
S i T i
S i T i
• Denominator: energy normalization term.
• is the time lag where the best match occurs.
xcorr = Time alignment problems
An alternative to xcorr (L_2-norm) between the two time series
2
1
( , ) ( ( ) ( ))
n
euclid
i
D S T S i T i
Euclidean distance
9. Euclidean Distance & xcorr
i
i
time
Euclidean distance:
aligns the i-th point on one time
series with the i-th point on the other
poor similarity score.
Correlation of well logs has always
been a labor-intense interactive
task. It is a pattern recognition
problem better solved by the
human eye than a computer.
Zoraster et al., 2004
We are trying to simulate
the procedure with the way
humans perform the
comparison.Elena Tsiporkova:
http://www.psb.ugent.be/.../DTWAlgorithm.ppt
10. Manual seismic-to-well tie
The forward model
Sonic log
P-wave
Vp
Well logs
Bulk
density
ρ
Acoustic
Impedance
I
1
1
i i
i
i i
I I
R
I I
Reflectivity
r
Computed
Statistical
Wavelet
Wavelet
w
Convolution output
Synthetic
s
15. How done manually
• Apply bulk shift and minimum amount of
stretching + squeezing to correlate major
reflectors
• QC – look at resulting interval velocity changes
16. Dynamic Time Warping?
i
i+2
i
i i
timetime
Euclidean distance:
aligns the i-th point on one time
series with the i-th point on the other
poor similarity score.
DTW: A non-linear (elastic) alignment:
produces a more intuitive similarity
measure.
It matches similar shapes even if they
are out of phase on the time axis.
A pattern matching technique that is
“visually perceptive and intuitive”
Elena Tsiporkova:
http://www.psb.ugent.be/cbd/papers/gentxwarper/DTWAlgorithm.ppt
17. Dynamic Time Warping?
Euclidean Distance
Sequences are aligned “one to one”
DTW
Nonlinear alignments are possible
Dr. Eamonn Keogh http://www.cs.ucr.edu/~eamonn/tutorials.html
18. How is DTW Calculated?
[Ratanamahatana, E. Keogh, 2005]
Every possible warping between two time series, is a path through
the matrix. We want the best one…
S
T 1
( , ) min
K
kk
DTW S T w K
T
Warping path w
S
This recursive function gives us the
minimum cost path
(i,j) = d(si,tj) + min{ (i-1,j-1), (i-1,j ), (i,j-1) }
[Berndt, Clifford, 1994]
19. How is DTW Calculated?
Synthetic
Trace
warping path
j = i – w
j = i + w
s1 s2 s3
t1
s4 s5 s6 s7
t2
t3
t4
t5
t6
t7
S_warped = s1 s2 s2 s3 s3
t1 t2 t3 t3 t4T_warped =
s4
t5
s5
t5
s6
t5
s7
t6
s7
t7
33. Discussion
• Pros and cons
– Independent of the selected window.
– Able to follow non linearities
– Only intended as a guide – not all stretching-
squeezing is realistic
– QC – examine changes in resulting interval
velocity curve
34. Conclusions
• DTW: optimal solution for matching similar events.
• DTW: complementary tool for seismic-to-well tie.
• Many other applications of DTW are possible for seismic data.
– log-to-log correlations, alignment of baseline and monitor
surveys in 4D seismics, PP and PS wavefield registration for
3C data.