(t,x) domain, pattern‐based ground roll removal

Abstract

We demonstrate the use of a new, t − x domain, pattern-based signal/noise separation technique to separate ground roll from primary reflection events. Ground roll is notoriously difficult to model with generality, but the technique requires a kinematically correct model of the noise. We obtain an imperfect model of the ground roll directly from the data itself, by application of a suitable lowpass filter. On a 2-D receiver line gather taken from a 3-D shot gather, in which the ground roll is spatially aliased and has nonlinear moveout, the separation results improve markedly over direct subtraction of the noise model from the data.

Multiple attenuation using at − xpattern‐based subtraction method

“…We refer to this approach as a "Wiener-like" method. It has been successfully used by Brown and Clapp (2000) for ground-roll attenuation and by Clapp and Brown (2000) for multiple separation.…”

“…A wellknown problem of the single convolution approach to modeling is that it models the kinematics for every multiple correctly, but does not preserve the relative amplitude of high-order multiples. Brown and Clapp (2000) show that a pattern-based approach can compensate for amplitude errors in the noise modeling.…”

“…For instance, Spitz (2000) uses f − x domain PEF while Brown and Clapp (2000) and Clapp and Brown (2000) use t − x domain PEF.…”

What can we do with a model of the multiples?

“…A first strategy might consist in subtracting this model from the data. A second strategy might consist in using this model as an estimate of the noise multivariate spectrum for a pattern-like attenuation of the multiples (Brown and Clapp, 2000). This multivariate spectrum can be reliably estimated with non-stationary prediction-error filters.…”

Multiple attenuation with multidimensional prediction‐error filters

“…Recently Spitz (1999) and Brown and Clapp (2000) presented a noise attenuation technique based on patterns. The main idea is simple: the noise and signal have different multidimensional spectra that PEFs can approximate (Claerbout, 1992) in order to perform the separation.…”