In this section: Coherent Noise, Random Noise, Swell Noise, De-Spiking, Acquisition Noise, Seismic Interference, Footprint Removal, Receiver Motion Correction.
See also: Refine (high frequency enhancement)
Spectrum offers a full range of methods to attenuate different types of seismic noise – the noise is classified into two categories – random noise and coherent noise. Some of the methods used to attenuate seismic noise are described below.
Coherent noise includes linear noise types, reverberations and multiples. Two types of coherent noise that require attention are guided waves and side scattered energy. Guided waves are trapped in a water layer or low velocity near surface layer and are dispersive – each frequency component propagates with a different velocity. Side scattered energy tends to have a large moveout range depending on the position of the scatterer acting as a point source with respect to the position of the recording cable.
Side scattered energy is usually observed on common shot gathers where they are identified by their varying moveout. They can also be observed as linear noise on stacked sections and on timeslices. Attenuation of energy associated with side scatterers is carried out in the F-K or tau-P domain. A linear event on a shot record will map to a radial line in the FK/FKK domain and to a point in the tau-P domain making them easy to filter/mute in either domain.
FK filters can be specified as polygons or fans and can be applied in either the X-T or FK domain and are specified in terms of F and K co-ordinates. The fans are defined by “pass” and “reject” slopes in ms per trace. The input data can be any multi-trace group such as shot records, receiver gathers or CDP gathers. The standard FK filter is a 2D filter – FKK filtering applies a full 3D filter – that is a function of frequency and two wave numbers; Kx and Ky.
Linear tau-p filtering can be used to attenuate noise by dividing the tau-p domain into pass and reject zones, similar to the FK filtering. The input data can be common shot, common receiver, common midpoint or common reflection point gathers. Spectrum’s Radon is resistant to spatial aliasing and can therefore reduce the need for trace interpolation before noise removal. Our radon algorithm also honours the true offset of the data, so that linear noise can be accurately attenuated even in irregularly sampled datasets.
Random noise can be categorised as noise in the spatial and temporal directions that is uncorrelated from trace to trace. It is usually stronger at late times rather than early times in recorded seismic data, and in general, filtering in a time variant manner can be used to attenuate most of this random noise. Another powerful process that attenuates much random noise is conventional CMP stacking, which significantly reduces the uncorrelated noise within the data. While time variant filtering may reduce the noise at later times, it does not necessarily attenuate the noise from trace to trace. One of the best methods to reduce random noise is based on spatial prediction filtering such as FX deconvolution.
FX deconvolution is used to attenuate spatially random noise by enhancing the spatially predictable components of the seismic trace spectra using a Wiener deconvolution in the FX domain. A prediction filter is calculated and applied to the data. Energy that is spatially predictable is passed and the un-predictable energy, classed as noise, is removed. The key to this method is the idea that reflection signals on the seismic are coherent, and the signal spectrum on any trace segment can be predicted. The aim of FX deconvolution is to provide faithful transmission of all signals, preservation of the signal character and removal of noise.
Swell noise is often reduced with an FX based attenuation routine – detection and repair of high amplitude windows of a trace by FX projection filtering. However Spectrum has developed a module, NOISERM, which performs frequency dependent noise removal that can achieve attenuation of swell noise without reducing low frequency signal.
Each pre-stack ensemble is transformed into frequency-time (FT) space using a STFT [Short Time Fourier Transform] algorithm. The transform is separated into amplitude and phase components for each frequency sub-band, and then the median spectral amplitude within each requested frequency sub-band is calculated for the ensemble.
Each sample within each frequency sub-band is compared against a median amplitude threshold – if the sample amplitude exceeds this threshold then it is replaced with a median amplitude value from its neighbouring traces.
The inverse STFT is then applied and the balanced ensemble is passed on. If the noise is isolated to a small set of frequency sub-bands or a subset of times, then the data which is analysed for noise removal can be limited by the user. The threshold determination and threshold violation search will then be constrained to only those sub-bands and times requested.
Spikes are impulsive noise bursts within the data and removal of them generally involves comparing amplitude values at a particular sample or trace with the neighbouring samples or traces. If the absolute amplitude values of the sample or trace under analysis, exceed those of the comparison sample or trace by a defined ratio, then the sample can either be zeroed or the amplitude value of the sample reset to a defined value.
Coherent noise also exists in land seismic datasets in the form of dispersive Rayleigh waves known as groundroll. This type of noise is characterised by having a low velocity, large amplitudes and low frequencies and can dominate the reflection energy in the recorded data. Attenuation of ground roll is usually carried out in the F-X domain where frequency dependent trace mixing is performed followed by horizontal correlation filtering at each frequency for the specified surface wave velocity. The filtered data is returned to the T-X domain and frequencies above a user defined cut-off are left unchanged.
Seismic interference is caused by other acquisition vessels or field crews shooting at the same time within the vicinity. The resulting noise observed on the seismic can be complicated especially since they are observed over a long distance and they can exhibit dispersion. A number of methods can be used to eliminate such seismic interference noise;
F-X attenuation applied in the domain in which the noise is randomized such as common offset domain. The noise trains are isolated and removed within the affected frequency bands. If the location of the noise source is known then tau-P attenuation can be used. The noise can be modelled and subtracted in the tau-P domain.
Footprint filtering works on a time slice basis. The time slices are typically obtained from stack data and converted to the Kx,Ky domain by means of a 2D Fourier transform. On a Kx,Ky amplitude spectrum of a time slice, the acquisition footprint shows itself as a repetitive pattern relating directly to the shot and receiver line spacing. Filters can be applied to the data in the Fourier domain and the filtered data transformed back to the x,y domain using an inverse 2D FFT.
Receiver Motion Correction
Receiver motion introduces a time-variant spatial shift into the data. Source motion converts the effects of the source signature from a single-channel convolution in time to a multi-channel convolution in time and space. For marine vibroseis acquisition both source and receiver motion is important. In 3D marine data receiver motion alone can produce significant artefact.