Filter: Seismic Frequency and Amplitude Removal, Signal-to-Noise Ratio, and WCSB Processing Workflows

A filter, in seismic processing, is any operation that removes undesirable portions of recorded data so that the geologic signal stands out more clearly against the noise that accompanies it, raising the signal-to-noise ratio of the dataset. Reflection seismic records are mixtures of useful energy, reflections from subsurface interfaces that carry information about depth, structure, and rock properties, and unwanted energy such as ground roll, air blast, wind noise, powerline hum, multiples, and random background that masks the reflections of interest. Filtering is the act of suppressing the unwanted content while preserving, as much as possible, the part of the wavefield that maps the subsurface. Filters are described by what they act on. A frequency filter, the most familiar kind, passes some band of frequencies and attenuates others: a low-cut or high-pass filter removes the low-frequency ground roll that dominates land records, a high-cut or low-pass filter removes high-frequency random noise, and a band-pass filter keeps only an intermediate band such as 8 to 80 Hz where the reflection energy lives. A notch filter removes a single narrow frequency such as 60 Hz electrical interference. Beyond simple frequency selection, processors apply filters that act on amplitude, on apparent velocity or dip, and on the relationship between traces. An f-k filter, working in the frequency-wavenumber domain, rejects energy by its dip and is the standard tool for removing coherent linear noise like ground roll and guided waves that share frequencies with the signal but travel at different apparent velocities. Deconvolution is a filter that compresses the seismic wavelet and suppresses short-period multiples to sharpen vertical resolution, while a deghosting filter removes the predictable ghost reflection from the surface. Filters can be applied trace by trace, across an ensemble, or to the whole volume, and they may be time-invariant or time-varying so that the pass band narrows with depth as high frequencies are lost to absorption. The central trade-off in all filtering is that no filter perfectly separates signal from noise; an overly aggressive band-pass that removes ground roll can also strip real low-frequency reflection energy needed for inversion, and a heavy high-cut smooths the section but destroys resolution. Processors therefore test filter parameters on representative gathers, inspect amplitude spectra before and after, and choose the mildest filter that achieves the interpretive goal. In the Western Canadian Sedimentary Basin, where targets range from shallow Mannville channels to deep Montney and Duvernay shale intervals, filtering is tuned to the bandwidth each target supports, and the same raw field record may be filtered differently for a structural interpretation than for a quantitative amplitude-versus-offset study where preserving true amplitudes outranks cosmetic noise removal.

Coherent Noise and the f-k Filter

The hardest noise to remove is coherent and shares frequencies with the signal, ground roll being the classic land example. A simple band-pass cannot reject it without also harming reflections, so processors transform the data into the frequency-wavenumber, or f-k, domain where energy separates by apparent velocity and dip. Ground roll, travelling slowly across the spread, maps to a distinct fan that can be muted while the steeper-moveout reflection energy is preserved, then the data is transformed back. The same approach in the tau-p (slant stack) domain targets linear noise and multiples. The cost is potential artifacts and aliasing if spatial sampling is coarse, so f-k filtering is parameter-tested on shot gathers before application.

Deconvolution as a Filter

Deconvolution is a filtering operation that reshapes the seismic wavelet rather than rejecting a frequency band. The recorded trace is the earth reflectivity convolved with a source wavelet plus reverberations; deconvolution designs an inverse operator that compresses that wavelet toward a spike and attenuates short-period multiples and ringing, sharpening vertical resolution and flattening the amplitude spectrum. Predictive deconvolution targets periodic multiples such as those from a high-impedance contrast, while spiking deconvolution broadens bandwidth. Because it whitens the spectrum, deconvolution must be balanced against noise amplification, and processors constrain it with a chosen operator length and white-noise factor tuned to the WCSB target interval.

Related Terms

Filtering only makes sense against its opposites: the signal is the reflection energy a filter tries to preserve, while noise is what it tries to remove, and the whole point is improving the ratio between them. Filters operate on data gathered during seismic acquisition and refined through deconvolution, itself a filtering step that compresses the wavelet. Because filters reshape amplitudes, they bear directly on amplitude versus offset analysis, where over-filtering can erase the very amplitude behaviour the interpreter is trying to measure.

Real-World WCSB Scenario

A producer shoots a 3D survey over a Duvernay light-oil play near Fox Creek, Alberta, targeting a shale interval near 3,300 m where the reflection bandwidth is modest and ground roll from the shallow section is severe. The processing contractor, on a budget near CAD 95,000 for the processing scope, tests filter parameters on a swath of shot gathers: an f-k filter to reject the slow-moving ground roll fan, a time-varying band-pass tightening from 10 to 95 Hz shallow toward 8 to 50 Hz at the Duvernay, and surface-consistent deconvolution to compress the wavelet.

The interpreter requires an amplitude-preserving product for an AVO screen, so the team documents every filter and keeps the low-cut gentle to retain the low-frequency energy the inversion needs. The result resolves the Duvernay and the overlying Ireton with usable amplitudes, and the AVO anomaly that survives the controlled filtering supports a horizontal well location that later tests commercial condensate rates.