Comparison of Spectral Decomposition Algorithms For Channel Detection in 3D Seismic Data
Comparison of Spectral Decomposition Algorithms For Channel Detection in 3D Seismic Data
Comparison of Spectral Decomposition Algorithms For Channel Detection in 3D Seismic Data
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/283214288
CITATIONS READS
0 195
3 authors, including:
Some of the authors of this publication are also working on these related projects:
Construction of the Lower Paleozoic extent’s maps biostratigraphy and analysis of tectonic evolution
of the marginal zone of the Eastern European Platform for estimation of unconventional hydrocarbon
deposits distribution View project
All content following this page was uploaded by Kamil Cichostępski on 26 October 2015.
Th SP2 03
Comparison of Spectral Decomposition
Algorithms for Channel Detection in 3D Seismic
Data
A. Kwietniak* (AGH - University of Science and Technology), K.
Cichostepski (AGH - University of Science and Technology) & M.
Kasperska (AGH - University of Science and Technology)
SUMMARY
The scope of the work was to chose the best spectral decomposition (SD) method for the channels
detection in the Miocene strata of the Carpathian Fordeep, SE Poland in the 3D seismic data volume. For
analysis three spectral decomposition algorithm were applied: based on Fast Fourier Transform (FFT),
Continuous Wavelet Transform (CWT) and based on complete ensemble empirical mode decomposition
(CEEMD). Because CWT results were considered rather poor the comparison was limited to FFT and
CEEMD. Additionally the results of instantaneous frequency were compared with the results of peak
frequency computed from CEEMD. Both algorithms of SD enabled to interpret channels, but the results
are slightly different, i.e. FFT show more coarse, linear structures, that are straightforward for channel
interpretation whereas CEEMD show more details and depicts where tuning effect occurs. So as to IF and
PF we value more instantaneous frequency due to much better results: it shows nicely linear structures and
is less noisy than PF. Moreover the computation of IF is much faster (‘on the fly’) than PF (couple of
hours for the whole volume).
The main goal of the work was to find the best possible method for determining the existence of
channels in Miocene sediments in the Carpathian Foredeep, SE Poland. The Carpathian Foredeep is
one of the main gas field in Poland, where the most common are structural gas traps, whereas the
potential of gas-bearing channels traps, that are associated with stratigraphic factors are not widely
studied. Hence, more and more interest is dedicated to channels detection in the seismic survey. We
applied two algorithms of spectral decomposition (SD): first based on Fast Fourier Transform (FFT)
and second which decompose the signal into a sum of oscillatory components, with guaranteed
positive and smoothly varying instantaneous frequencies (Han and van der Baan, 2013): complete
ensemble empirical mode decomposition (CEEMD). Additionally, we computed instantaneous
frequency (IF) and instantaneous phase. We found differences between results of SD.
Method
The main aim of the work was to determine which algorithm of decomposition give better results in
the study area. Particularly we wanted to answer the question whether CEEMD algorithm has
significant advantages (Han and van der Baan, 2013) over other decomposition algorithms. The
chosen methodology was inspired by the study of channels detection first suggested in the work of
Partyka (Partyka, 1999; Chopra and Marfurt, 2007) for the 3D seismic survey from the Gulf of
Mexico. The steps of our analysis were as follows:
1. Decomposition based on CEEMD algorithm was computed in Hampson-Russell for the whole
3D volume. The output were iso-frequency slices. We analysed closely following frequencies:
30, 35, 40, 45 Hz. Additionally this algorithm enables to compute peak frequency (PF) of the
decomposed volume, which can be compared with the results of instantaneous frequency.
2. FFT algorithm was performed in OpendTect. Basing on the well M-1 we found the optimal
length of the window used for spectral decomposition for a problem at hand. Then we
computed SD with the window length of 40 ms with step of 5 Hz. We computed slices of
frequency with the same pattern as for CEEMD decomposition. During analysis we tested
also Continuous Wavelet Transform algorithm, but the results were not satisfactory and we
decided not to include them in the main comparison.
3. Instantaneous frequency and instantaneous phase were computed in Petrel.
4. All results were normalized in scale and compared in Petrel. In this abstract, due to limited
space, we will show only horizontal slices examples.
Examples
In Fig. 1 the results of two algorithms are shown (Fig.1a CEEMD, Fig.1b FFT). We compare
horizontal slices for frequency slice of 35 Hz. It is visible that channels are present and can be
recognised in both methods. However, FFT characterizes with more linear structures, whereas
CEEMD shows more details that are not visible in FFT. We link this behaviour to the length of the
time window used for FFT decomposition process. In Fig. 2 we compare results of instantaneous
frequency and peak frequency computed after spectral decomposition. The channels are visible in
both figures (A and B) but in our opinion IF shows better results: the interpreted channels show
constant alignment and liner character in comparison with PF.
Conclusions
The analysis enable us to state that: the new algorithm of spectral decomposition for this particular
kind of data does not show significant improvements in comparison with FFT method. The results are
similar, although we price more FFT algorithm. Its results are more continuous and show nice
lineaments that we interpret as channels. We found out that different frequency slices revealed
features that we interpret as parts of channels that characterizes with slightly different geometry, i.e.
features closely related to the geometrical scale of the channel. Results of instantaneous frequency as
well as peak frequency nicely depict the channel geometry. We found the results of instantaneous
Acknowledgements
The work was part of the research project no. AGH 18.18.140.252. The authors want to express their
deepest gratitude to prof. Kaja Pietsch for her valuable suggestions and support. Authors were
sponsored by the AGH dean’s grant of numbers: 15.11.140.345,15.11.140.336 and 15.11.140.541.
References
Chopra, S. and Marfurt, K.J [2007] Seismic Attributes for prospect identification and reservoir
characterization. SEG Geophysical Developments Series, No. 11. 138-141,
Han, J. and van der Baan M. [2013] Empirical mode decomposition for seismic time-frequency
analysis. Geophysics, 78(2), March-April 2013, p. O9-O19,
Partyka, G., Gridley, G.J. and Lopez, J. [1999] Interpretational applications of spectral decomposition
in reservoir characterization. The Leading Edge, March 1999, 353-360.