Ground-roll noise suppression by combining the local radial trace and mean filtering

Reflection seismic data is often contaminated by a variety of coherent and incoherent noises. Ground-roll is a type of surface wave. It is a form of coherent linear noise, which propagates at the surface of the earth, at low velocity and low frequency. This kind of noise will often obscure the seismic reflection data.
Various methods have been introduced by many authors to attenuate the ground-roll in reflection seismic data. Most of these methods are based on a transform. Deighan and Watts (1997) tried to eliminate the ground-roll by using the wavelet transform. A number of researchers used Karhunen-Loeve transform to suppress the ground-roll noise from seismic data (Liu, 1999; Montagne and Vasconcelos, 2006). Porsani et al. (2009) presented the singular value decomposition (SVD) filtering method for attenuation of the ground-roll. In this method, the seismic data must first be NMO corrected. Naghizadeh and Sacchi (2011) eliminated the ground-roll noise by scale and direction guided curvelet transform.
    One of the most important and common methods to attenuate coherent linear noise such as ground-roll in a prestack gather is based on the radial trace transform. It is a simple coordinate transform of offset-time (X-T) into the radial-time (R-T) coordinate. This transform was introduced and developed in the Stanford Exploration Project by Clearbout (1975), Ottolini (1979, 1982) and Taner (1980). They used the radial trace transform for migration of seismic data and multiple attenuation. Clearbout (1983) and Henley (1999, 2000 and 2003) attenuated the coherent linear noise by the radial trace transform.
We can separate the coherent linear noise from reflection seismic data when the transform coordinate paths are appropriately aligned with the coherent noise direction in the X-T domain. Applying the radial trace transform on a seismic shot gather causes coherent linear noise which is distributed across many traces in the X-T domain maps into relatively few radial traces and reduces their frequency. The most straightforward technique to reduce coherent noise in the R-T domain is to apply a high-pass (low-cut) filter to the radial traces. In practice, interpolation is an important stage in the radial transform and can leave a damaging impact on performance. Therefore, application of the transform will result in missing data. However, advanced interpolation methods can reduce the omitted data problem.
    In this study, we used a different method based on radial trace filtering, named local radial trace mean filtering that does not require a forward/inverse radial trace transform. Consecquently, it does not require interpolation algorithms. To investigate the efficiency of the proposed method, we test the local radial trace mean filtering on both synthetic and real seismic data. We also compare the obtained results by those from the f-k filtering for seismic data ground-roll suppression. To investigate further, we apply the method on synthetic seismic contaminated by random noise and find that the method is not sensitive to the presence of random noise. Also, in cases where the receiver intervals are not the same, unlike the f-k filtering, local radial trace mean filtering is very effective. Therefore, we can say that the local radial trace mean filtering method is a suitable tool for coherent linear noise attenuation, especially the ground-roll.