coherence filtering

Coherence filtering (also called structure-oriented filtering or dip-steered filtering) in seismic data processing is a noise-attenuation technique that applies a smoothing filter oriented along the local dip and azimuth of each reflection surface in a 3D seismic volume, exploiting the lateral continuity of genuine reflections to discriminate them from random noise, which lacks spatial coherence across adjacent traces; the filter is constructed by first estimating the 3D dip field at every sample point using the gradient structure tensor, plane-wave destruction filters, or semblance-based dip scanning over an aperture of 3x3 to 7x7 traces and 8 to 20 ms, then applying a mean, median, or Gaussian-weighted average of samples along the estimated dip surface to improve signal-to-noise ratio while preserving sharp discontinuities at faults and stratigraphic edges where the dip field changes abruptly. Coherence filtering differs from conventional time-domain or frequency-domain methods (band-pass filtering, f-x deconvolution, horizontal median filtering) in adapting to geological structure: a horizontal median filter on a 30-degree dipping reflector smears the reflection peak across time samples, while a dip-steered filter along the same surface preserves peak shape while averaging only the orthogonal noise component. In WCSB 3D seismic processing, coherence filtering is applied post-stack before attribute extraction because attribute computation amplifies noise: at WCSB target depths of 1,000 to 4,000 m, random noise levels of 5 to 20 percent of reflection amplitude produce coherence anomalies of 0.05 to 0.15 units that mimic geological discontinuities if not first attenuated, generating false fault interpretations in WCSB structural mapping and erroneous channel delineation in Mannville and Viking programs; dip-steered filtering is implemented in SLB Omega/Delphi processing systems and in Petrel, Kingdom, and OpendTect interpretation platforms on WCSB 3D datasets of 100 to 5,000 km2.

• Dip estimation methods underpinning coherence filtering in WCSB 3D seismic processing: The accuracy of coherence filtering depends critically on the quality of the estimated dip field used to orient the smoothing operator; three dip estimation methods are used in WCSB processing: gradient structure tensor (GST) methods compute the spatial gradients of the seismic amplitude field and derive the dominant dip from the eigenstructure of the gradient tensor matrix averaged over a local analysis window (computationally efficient, applicable to WCSB datasets of 500 to 5,000 km2 in batch processing, but sensitive to noise in areas of low signal-to-noise ratio); plane-wave destruction (PWD) filter methods estimate dip by finding the dip value that minimizes prediction error between adjacent traces using a recursive causal filter (more accurate in low-SNR areas than GST but computationally intensive); and semblance scanning methods compute dip by maximizing the multi-trace semblance (coherence) over a range of candidate dip values at each sample (highest accuracy for complex structures including WCSB Foothills thrust-fold belts where dips exceed 30 to 60 degrees and GST methods fail). WCSB Alberta Plains datasets (gentle dips of 1 to 5 degrees) use GST dip estimation as the standard method, while WCSB Foothills and Deep Basin programs at Rocky Mountain Trend and Grande Prairie use PWD or semblance scanning to handle the steep and rapidly varying dip field of thrust-imbricated Cretaceous and Jurassic formations at depths of 2,000 to 6,000 m.
• Edge-preserving variants of coherence filtering for WCSB fault and channel imaging: Standard dip-steered coherence filters that average all samples along the estimated dip surface produce excellent noise attenuation on continuous reflectors but smear the amplitude discontinuities at fault planes and channel margins that are the primary targets of WCSB structural and stratigraphic interpretation; edge-preserving coherence filters address this by using the local coherence (similarity) value as a weight in the smoothing operator, reducing the smoothing weight at locations where the coherence is low (discontinuity present) and applying full smoothing where coherence is high (continuous reflection). Three edge-preserving variants are used in WCSB 3D interpretation: the bilateral filter (combining a Gaussian spatial weight with a range weight based on amplitude similarity, preserving edges where amplitude changes abruptly); the non-local means filter (finding similar patches elsewhere in the volume rather than only in the local neighborhood, effective for WCSB data with spatially repetitive geology such as regularly spaced Viking channels); and the anisotropic diffusion filter (applying the diffusion equation with a diffusion tensor oriented along the dip surface, stopping diffusion perpendicular to edges). In WCSB Nisku carbonate reef interpretation at Pembina and Brazeau fields, edge-preserving coherence filtering has improved fault detection sensitivity by 20 to 35 percent (measured by the number of interpreted fault segments identified by expert interpreters before and after filtering) while maintaining reef-flank amplitude character used for porosity prediction.
• Coherence filtering parameters and quality control in WCSB 3D seismic attribute workflows: Coherence filtering parameter selection for WCSB 3D datasets balances noise attenuation against resolution: the filter aperture (number of traces and time samples used in the smoothing window) controls the trade-off between noise rejection (larger aperture, more averaging) and lateral resolution (smaller aperture, finer structural detail preserved); for WCSB Mannville channel characterization where channel widths of 200 to 800 m must be resolved within a 3D survey with 25 m trace spacing, a filter aperture of 5x5 traces (125 m) applied 2 to 3 times is preferred over a single-pass 9x9 trace aperture (225 m) that would smear channel margins. Quality control of coherence filtering in WCSB processing uses frequency-wavenumber (f-k) spectra before and after filtering (the filter should attenuate noise in the high-wavenumber, low-frequency quadrant without removing high-frequency, high-wavenumber signal from faults and channels), amplitude spectrum comparison (filter should not reduce reflection amplitudes below a threshold of 5 percent relative to unfiltered data at the dominant frequency), and blind-well tests where filtered data is used to predict lithology at a known well location and the prediction accuracy is compared to unfiltered data prediction. Over-filtering (too many passes or too large an aperture) is a common quality control failure in WCSB interpretation workflows that produces geologically unrealistic smooth coherence maps lacking genuine fault detail.
• Coherence filtering enabling seismic attribute extraction for WCSB unconventional resource characterization: In WCSB Montney siltstone, Duvernay shale, and Muskwa-Otter Park Formation horizontal well programs, coherence filtering applied before spectral decomposition and curvature attribute extraction is critical for distinguishing natural fracture corridors (genuine geological discontinuities with coherence lows at 50 to 200 m spacing) from noise-induced false discontinuities (appearing at trace spacing, 12.5 to 25 m, with no geological significance); without dip-steered pre-conditioning, spectral decomposition frequency slices for WCSB Montney at the Groundbirch, Dawson, and Progress fields show a speckled noise pattern that obscures the 100 to 500 m scale channel and fracture features that govern production heterogeneity in Montney horizontal wells. Coherence filtering before AVO (amplitude versus offset) attribute extraction is also applied in WCSB Cardium tight oil and Glauconitic sandstone programs where AVO gradient attributes (used to distinguish gas from oil-saturated sandstone) are computed from pre-stack angle gathers; dip-steered filtering of the angle gathers before AVO extraction reduces the noise-induced scatter in cross-plots of AVO intercept versus gradient that otherwise makes fluid discrimination ambiguous at WCSB Cardium seismic signal-to-noise ratios of 4 to 8 dB in the far offset angle stack.
• Coherence filtering in WCSB time-lapse (4D) seismic processing for SAGD steam chamber monitoring: Time-lapse seismic monitoring of WCSB SAGD steam chambers in Athabasca oil sands (repeated 3D surveys over the same area acquired 1 to 3 years apart) requires careful coherence filtering to separate the genuine 4D signal (amplitude and time-shift changes caused by steam replacing cold bitumen) from acquisition and processing repeatability noise; the 4D noise level in WCSB SAGD monitoring surveys (NRMS, normalized root mean square difference between repeated surveys, typically 15 to 35 percent of the signal amplitude) is dominated by near-surface variability and receiver coupling differences that appear as incoherent noise in the 4D difference volume. Dip-steered coherence filtering applied to both the baseline and monitor surveys before 4D subtraction improves the 4D signal-to-noise ratio by 20 to 40 percent (measured by NRMS reduction and improvement in the signal detection threshold for steam chamber boundaries) at WCSB SAGD monitoring programs operated by Cenovus at Foster Creek, MEG Energy at Christina Lake, and Canadian Natural Resources at Kirby; the filter must use identical dip fields estimated from a merged average of the baseline and monitor surveys to avoid introducing artificial 4D anomalies from filter-induced amplitude differences at locations where the dip estimate changes between surveys.

Coherence Filtering Improving WCSB Viking Channel Delineation

A WCSB operator interpreting a 220 km2 3D seismic dataset over a Viking Formation tight oil play in central Alberta found that raw-stack coherence attribute maps showed extensive salt-and-pepper noise artifacts at the 12.5 m trace spacing level that made channel boundary identification ambiguous over 40 percent of the survey area. A dip-steered coherence filter using GST dip estimation and a 5x5 trace, 12 ms bilateral edge-preserving filter applied 3 times reduced the NRMS noise level from 22 percent to 8 percent in the Viking Formation time window. Post-filter coherence maps resolved two additional meandering channel systems not visible in the pre-filter volume, covering 18 km2 of productive Viking sandstone. Well control at 8 locations confirmed the channel geometry predicted by the filtered coherence map within 50 m in 7 of 8 cases. The horizontal well program targeting the two new channels (4 wells, 1,400 m lateral length each) achieved an average 12-month production of 3,200 barrels of oil equivalent per well, compared to 2,100 barrels for the pre-filter-guided wells drilled in the first phase of the program, a 52 percent improvement attributed to more accurate channel steering enabled by the coherence filtering workflow.

Related Terms

Coherence is both the attribute coherence filtering enhances and the measure used in edge-preserving filter variants to weight smoothing; dip-steered filtering improves coherence attribute maps used for WCSB fault and channel interpretation. Seismic attribute extraction in WCSB 3D programs requires coherence filtering as a pre-conditioning step; without dip-steered smoothing, noise-induced false anomalies in coherence, curvature, and spectral decomposition lead to misinterpretation of structural and stratigraphic features. Dip estimation is the prerequisite for coherence filtering; accuracy of GST, PWD, or semblance-scanned dip fields controls filter effectiveness in WCSB complex-structure areas. Time-lapse seismic (4D) WCSB SAGD monitoring requires identical coherence filtering of baseline and monitor surveys before subtraction; different dip fields for each vintage create false 4D anomalies. Structure-oriented filtering is a synonym for coherence filtering; both describe dip-steered noise attenuation along geological reflection surfaces rather than horizontal time slices.