Figure Caption. Identification of shallow horizon and verification of dipping horizon integrity. Data were collected in 1994 with a 4.0 ms sampling rate, 25 m channel spacing (spread=6 km), and a 200 m near offset. In all cases, shot 3 is on the left and shot 4 is on the right. Base sections in (A) are resampled to 8.0 ms, bandpass filtered (passing 6-45 Hz, cosine tapering 2-6 and 45-55 Hz) to remove some ground roll, and AGC'd with a 4.0 s window. Boxes indicate locations of Figures B, C, and D. A close-up of the first 2.0 s and 120 channels is shown in (B) where the top mutes have now been applied. Additional processing in (B) includes: elevation and residual (Ronen and Claerbout, 1985) statics, NMO, and trace equalization based on mean amplitudes from 8 s to 9 s. Unmuted linear noise, presumably related to refractions, is marked with arrows labeled "F", other arrow labels indicate noise, "NZ", such as ground roll and randomly occurring noise bursts, whereas reflections are marked with an "R". In (C) and (D), the FK gather CC filter has been applied to (B) and (A), respectively. FK gather CC filter parameters are: window dimensions of 300 ms with 300 ms overlap by 120 traces with no overlap, and correlation coefficient pass range is from 0.9 to 1.0. A reflection is apparent at 1.0 s in shot 3 of (B), yet undetected in shot 4; however, after FK gather CC filtering, the event is clear in (C). Shallower events are also recognizable after filtering in (C). The dipping reflections in (D) are improved by FK gather CC filtering, rather than being damaged. Frequency content is preserved by avoiding heavy low-cut filtering and opting for CC filtering to remove ground roll.

Access FTP Cite to Download Software

Noise Reduction In Seismic Data Using Fourier Correlation Coefficient Filtering*

Doug Alsdorf
Cornell University, Institute for the Study of the Continents, Dept. of Geological Sci., Ithaca, NY

* Published in Geophysics, v62, no. 5, pp. 1617-1627, 1997.

Abstract. The correlation coefficient between two frequency (or two wavenumber) components equals the cosine of their phase angle difference. This relation can be exploited to build a filter that separates noise from signal in seismic data in either the FX or FK domain (termed "correlation coefficient filtering"). To implement this filter, seismic data are first divided to form two subsets which are then compared using the cosine function. Signal is defined as the correlative frequencies (or wavelengths) while non-correlative energy is attributed to noise. Depending on the application, appropriate subsets may consist of (1) groups of adjacent traces or (2) low-fold stacks created from differing shot gathers. When comparing adjacent traces (i.e., #1), the correlation coefficient filter combines both phase and dip information and assumes that reflections advance relatively little in time across traces, and less than the noise. Correlation coefficient filtering of low-fold stacks (i.e., #2) does not depend on dip; reflections are assumed to be present in both subsets whereas the noise is found only in one data set. Hence, the reflections are correlative and the noise is non-correlative. In either case, the filter reduces linearly dipping coherent energy, ground roll, and randomly occurring noise bursts while generally maintaining signal integrity. A primary advantage of this filter is its simplicity. It is implemented much like a simple bandpass filter, thus requiring much less parameterization than alternative noise-reduction methods.

Return to Doug's Home Page