|
<< Click to Display Table of Contents >> Navigation: Filters > Despike GCube |
Despike GCube removes high-amplitude spectral spikes from a 3D post-stack seismic cube stored on disk as a SEG-Y file. Unlike trace-by-trace despiking, this module operates in the frequency domain across a three-dimensional sliding window, comparing each trace's spectral energy at every frequency against the energy distribution of its spatial neighbors. This cross-volume comparison makes the module effective at identifying and attenuating impulsive noise that is spatially isolated, such as single-trace amplitude bursts or narrow-band coherent interference that deviates strongly from the surrounding background.
The processing works as follows: for each time position in the volume, an overlapping Hamming-tapered window of samples is extracted and Fourier-transformed for every trace in the cube. At each frequency, the spectral amplitude of the central trace is compared against the amplitude distribution of all traces within the configured inline and crossline half-windows. If the central trace's amplitude exceeds the threshold derived from the neighborhood distribution, the spectral coefficient is scaled back to 80% of the local median energy before being inverse-transformed. The windowed results are recombined using an overlap-add scheme to reconstruct the full-length output cube. The cleaned cube is written to a new SEG-Y file on disk.
Note: This module is deprecated. For new projects, use the standard Despike or De-spike Distributed modules which operate directly on data connectors without requiring intermediate SEG-Y files on disk.
This module reads data directly from a SEG-Y file on disk rather than from a pipeline data connector. The input and output file paths are specified as parameters (see below). No data connector inputs are required.
The full path to the input 3D seismic cube in SEG-Y format (.sgy or .segy). The file must be a regular 3D post-stack volume arranged as inline × crossline × time. The module reads the entire cube into memory before processing, so ensure the file is accessible and that sufficient memory is available for the dataset size.
The full path to the output SEG-Y file (.sgy or .segy) where the despiked cube will be written. The file will be created or overwritten. The output cube has the same dimensions, sample interval, and geometry as the input; only the amplitude values are modified where spikes were detected.
The half-width of the spatial analysis window in the inline direction, measured in traces. The full inline window spans 2 × Trace window InLine + 1 traces and defines the neighborhood from which the reference spectral energy distribution is computed for each central trace. Default value: 10 traces. Increase this value in areas with sparse spike occurrence to obtain a more stable reference distribution. Reduce it when the noise is spatially correlated over large areas and you want to preserve subtle lateral amplitude variations. If the specified window exceeds half the cube dimension in the inline direction, it is automatically clamped to avoid boundary errors.
The half-width of the spatial analysis window in the crossline direction, measured in traces. The full crossline window spans 2 × Trace window XLine + 1 traces. Default value: 10 traces. For volumes with strongly anisotropic trace spacing (where inline and crossline bin sizes differ significantly), you may wish to set different values for inline and crossline windows so that both windows cover a similar physical aperture on the ground.
The half-length of the time analysis window in samples. At each time position, a segment of 2 × Sample window + 1 samples is extracted, tapered with a Hamming window, and Fourier-transformed to produce the spectral representation used for spike detection. Default value: 40 samples (minimum: 10 samples). A larger sample window provides better frequency resolution, which helps distinguish narrow-band spike energy from broadband signal energy, but increases computation time. A smaller window improves time resolution but reduces the ability to characterize spectral content. The sample window must be smaller than the total number of samples in the cube; otherwise the module will report an error.
The step size in samples by which the analysis window is advanced through the trace for each successive overlap-add iteration. Default value: 10 samples (minimum: 1 sample). This parameter controls the overlap between successive analysis windows. A smaller step produces more overlap and smoother reconstruction at the cost of longer processing time. A larger step reduces computation time but may introduce subtle amplitude discontinuities at window boundaries. This value must always be less than 2 × Sample window + 1; if it equals or exceeds that limit the module will report an error. If an even number is provided, it is automatically incremented by one to keep the window centered.
A dimensionless multiplier that determines how much a trace's spectral energy must exceed the neighborhood reference level before it is classified as a spike. Default value: 3.0 (minimum: 1.0). When the Threshold criterion is set to Median or Lower quartile, a spike is detected when the central trace energy exceeds the sorted neighborhood median multiplied by this threshold value. Lower values (closer to 1.0) are more aggressive and will suppress more energy, potentially affecting genuine reflections; higher values are more conservative and only remove the strongest anomalies. When using the Regression criterion, the threshold is not applied directly — spike detection is based on a fixed 3-sigma statistical criterion derived from the neighborhood energy distribution, and this parameter is ignored.
Selects the statistical method used to establish the reference energy level against which each trace is compared. Default: Median. Three options are available:
Median — The neighborhood spectral amplitudes are sorted and the middle value is taken as the reference energy. A trace is flagged as a spike if its amplitude exceeds this median multiplied by the User-supplied threshold. This is the most robust option for data with occasional large-amplitude outliers and is the recommended default for most post-stack cubes.
Lower quartile — Similar to the Median criterion but uses the lower half of the sorted neighborhood distribution as the reference. This makes the detection more sensitive to high-energy outliers because the reference is biased toward the quieter traces in the neighborhood. Use this option when spikes occupy a non-negligible fraction of the spatial window and might otherwise bias the median upward.
Regression — The mean and standard deviation of the neighborhood spectral energies are computed, and a spike is detected when the central trace energy exceeds the mean plus three standard deviations (mean + 3σ). The User-supplied threshold parameter has no effect in this mode. This criterion is appropriate when the background energy follows an approximately Gaussian distribution and you want an objectively defined statistical cutoff rather than a user-tuned multiplier.