A pattern-based technique for ground-roll and multiple attenuation

We present a pattern-based method that separates coherent noise from signal. This method finds its mathematical foundation in the work conducted by Nemeth (1996) on coherent noise attenuation by least-squares migration. We show that a similar inverse problem can be formulated to attenuate coherent noise in seismic data. In this paper, we use deconvolution with prediction error filters to model the signal and noise vectors in a least-squares sense. This new formulation of the noise separation problem has been tested on 2-D real data for ground-roll and multiple attenuations. So far, it achieves similar results to the approach used by Brown and Clapp (2000) and Clapp and Brown (2000). However, we show that the main strength of this new method is its ability to incorporate regularization in the inverse problem in order to decrease the correlation effects between noise and signal.