Академический Документы
Профессиональный Документы
Культура Документы
ccnorm $ % i ai bi
"
%a %b
i 2 i
2 i i
is used, where ai and bi are time series in the local windows around the current sample for two adjacent traces, shifted by a given time lag. For each sample, the maximum of the local correlation with the previous trace, representing the trace-to-trace event time shift, is automatically selected. If required, the resulting time-shift panel can then be cleaned by the application of a moving-window two-dimensional median filter. This filtering step is crucial in practice. It accounts for outliers in the time-shift panel due to noise or local event discontinuity in the input gather. In a second step the information in the (cleaned) time-shift panel is used to calculate moveout-corrected traveltimes t0 for all samples in the input gather. This is done recursively, trace by trace, starting from the nearest-offset trace (assumed to represent zero-offset traveltimes). The moveout-corrected traveltime values t0 are used to map the amplitude of each sample to its correct traveltime location in the moveout-corrected gather using an appropriate interpolation (sinc-interpolation).
3036
Random noise removal and coherency enhancement The automatic moveout procedure described in the previous section breaks down if the signal-to-noise ratio in the CMP gather is low and events can not be correlated from trace to trace. In such situations it may be necessary to enhance the coherency and continuity of reflection events and remove random noise (it is assumed that multiple reflections have already been removed). For that purpose we apply a noise removal and coherency enhancement algorithm based on local high-resolution linear Radon transforms. Assuming that reflection events, when restricted to a small offset range, can be regarded as locally linear, it should be possible to focus such events in the local linear Radon domain. Random noise, on the other hand, cannot be easily described with a small number of slopes. Imposing a strong sparseness constraint during an inversion-type Radon transform should therefore enhance coherent events and suppress random noise (e.g., Thorson and Claerbout, 1985). In order to improve the efficiency of the Radon transform, we use a frequency-domain algorithm. As frequency domain high-resolution Radon transforms do not allow to apply a sparseness constraint along the time axis, focusing problems and artifacts may appear. To avoid these, we apply the transform in overlapping time windows. For the sparseness constraint we combine the approach of Herrmann et al. (2000) with iterative approaches as described, e.g., by Sacchi and Ulrych (1995). The involved linear equation systems are solved by iterative conjugategradient (CG)-type methods, the number of iterations determining the strength of the sparseness constraint (e.g., Trad et al. 2003). In the context of random noise removal, a very small number of CG-type iterations is used. In practice, a symmetric offset window is defined around each trace of the input gather. Within this offset window, overlapping time windows are defined and for each of these time windows, the high-resolution linear Radon transform, as described above, is performed, using the considered central trace as reference. The linear interpolation of these transform results along the time axis gives the global Radon transform result for the considered offset window. The central trace is reconstructed by linear horizontal stacking of the corresponding Radon traces. If a weak sparseness constraint is used during the transform, a perfect reconstruction is obtained, while a strong sparseness constraint results in a reconstructed trace that only contains the most coherent events (Figure 2). A mute in the Radon domain before inverse transforming can be used to suppress selected coherent events (e.g., linear noise).
a)
b)
c)
d)
Figure 1: Automatic moveout correction based on local correlations, demonstrated on a noise-free synthetic data example. a) Input gather. b) Result of automatic moveout correction. c) Time shifts (in seconds) determined from local correlations. d) t0 values (in seconds) computed from the time shifts.
The t0 panel contains information on the kinematics of all reflection events in the input gather and can be used for other purposes than the moveout correction itself. The underlying assumption is that events in the gather are not crossing or interfering, so that a unique moveout-corrected traveltime value can be assigned to each sample. The application of the automatic local-correlation-based moveout correction is illustrated on a noise-free synthetic data example in Figure 1. The time-shift panel in Figure 1c is obtained from the input gather in Figure 1a by local correlations as described above. Although not necessary in this example, a slight filtering of the time-shift panel has been applied. The t0 values calculated recursively (from small to large offsets) from the time shifts are shown in Figure 1d. These are used to remap the amplitudes of the input gather into their corresponding moveout-corrected positions. The result (after low-pass filtering) is shown in Figure 1b. As expected, the moveout correction process breaks down where events are interfering. However, where interference is not a problem, the moveout-corrected gather does not suffer from wavelet stretch.
3037
a)
b)
c)
Figure 2: Coherency enhancement by local linear Radon transforms with a strong sparseness constraint. a) part of a gather with random noise. b) 3 local linear Radon panels, corresponding to the 3 traces in the center of Figure 2a (using 7 traces per window), with a strong sparseness constraint. c) Result of Radonbased noise removal.
Procedure Based on the coherency enhancement and automatic moveout correction methods presented above, the procedure for obtaining a moveout-corrected gather involves the following steps: Firstly, the local linear highresolution Radon transform with strong sparseness constraint is applied to remove random noise. If required, more than one pass of noise removal can be performed and if required, coherent noise can also be suppressed in the local linear Radon domain. Secondly, the automatic local event correlation procedure is applied to the cleaned (coherency-enhanced) gather to obtain a time-shift panel and a t0 panel as described above. The t0 panel is then used to perform a moveout correction either of the coherencyenhanced gather or of the original gather. If the coherency-enhanced gather is used only for determining the time-shift and t0 panels, while the moveout correction itself is applied only to the original gather, other methods not necessarily amplitude-preserving can be used for coherency/continuity enhancement. Real data example To demonstrate the entire procedure of noise removal and automatic moveout correction, it is applied to the CMP gather in Figure 3. The gather contains reflection events, linear noise, random noise and dead traces. As a first step the noise-removal technique based on local high-resolution Radon transforms, as described above, is applied. The resulting coherency-enhanced gather is shown in Figure 4a along with the removed noise (Figure 4b). The dip decomposition obtained with the local linear Radon transform allows suppressing the steeply dipping linear coherent noise visible in Figure 4a by simply removing the corresponding Radon-domain traces. The gather after linear
Figure 3: Real data example: CMP gather used to demonstrate the noise removal and automatic moveout correction procedure (data courtesy of Statoil).
Conclusions We have demonstrated the application of an automatic moveout correction scheme based on local event correlations on a noisy real data example. To stabilize the process, event coherency in the input gather is increased by applying a noise removal scheme based on high-resolution local linear Radon transforms. The moveout correction procedure itself is stabilized by filtering intermediate results of the process (time-shift fields). In general, the described automatic moveout correction procedure depends on event continuity in the data and breaks down if events are interfering or multiples have not been removed. Also, it is assumed that the near-offset trace is close to zero offset. The presented noise removal scheme should be of interest beyond the context of automatic moveout correction.
3038
a)
b)
a)
b)
c)
d)
c)
d)
Figure 4: a) Gather after one pass of Radon-based noise removal. b) Removed noise. c) Gather after removal of linear noise in the local Radon domain and a second pass of random noise suppression. d) Total removed noise.
Figure 6: a) Result of the automatic moveout correction of the gather in Figure 4c, using the t0 values of Figure 5b. b) Automatic moveout correction of the original gather (Figure 3) using the t0 values of Figure 5b. c) The moveout-corrected gather of Figure 6b after horizontal median filtering (window length: 9 traces). d) Noise removed by horizontal median filtering.
References Herrmann, P. Mojesky, T. Magesan, M. and Hugonnet, P., 2000, De-aliased, high-resolution Radon transforms, 70th Ann. Internat. Mtg., Soc. Expl. Geophys., Expanded Abstracts, 1953-1956 Sacchi, M. D. and Ulrych, T. J., 1995, High-resolution velocity gathers and offset space reconstruction, Geophysics, 60, 1169-1177 a) b)
Figure 5: Intermediate results of automatic moveout correction of the gather in Figure 4c: a) Time shifts (seconds) determined from local correlations, after smoothing with a two-dimensional median filter. b) t0 values (seconds) computed from the time shifts.
Taner, M. T. and Koehler, F., 1969, Velocity spectra digital computer derivation and applications of velocity functions, Geophysics, 34, 859-881 Thorson, J. R. and Claerbout, J. F., 1985, Velocity stack and slant stack stochastic inversion, Geophysics, 50, 27272741 Trad, D., Ulrych, T. and Sacchi, M., 2003, Latest views of the sparse Radon transform, Geophysics, 68, 386-399
Acknowledgments This work has been supported by the Norwegian Research Council/PETROMAKS. We thank Statoil for permission to show the data example.
3039
EDITED REFERENCES Note: This reference list is a copy-edited version of the reference list submitted by the author. Reference lists for the 2006 SEG Technical Program Expanded Abstracts have been copy edited so that references provided with the online metadata for each paper will achieve a high degree of linking to cited sources that appear on the Web. REFERENCES Herrmann, P., T. Mojesky, M. Magesan, and P. Hugonnet, 2000, De-aliased, highresolution radon transforms: 70th Annual International Meeting, SEG, Expanded Abstracts, 19531956. Sacchi, M. D., and T. J. Ulrych, 1995, High-resolution velocity gathers and offset space reconstruction: Geophysics, 60, 11691177. Taner, M. T., and F. Koehler, 1969, Velocity spectra - Digital computer derivation and applications of velocity functions: Geophysics, 34, 859881. Thorson, J. R., and J. F. Claerbout, 1985, Velocity stack and slant stochastic inversion: Geophysics, 50, 27272741. Trad, D., T. Ulrych, and M. Sacchi, 2003, Latest views of the sparse Radon transform: Geophysics, 68, 386399.
3040