We propose a novel method for random noise attenuation in seismic data by applying nonstationary autoregression (NAR) in frequency-space (f-x) domain. Nonstationary autoregression can adaptively predict seismic events of which slopes vary in space. The key idea of this abstract is to overcome the assumption of linearity and stationarity in f-x deconvolution technique. The conventional f-x deconvolution uses short temporal and spatial analysis windows to cope with the nonstationary of the seismic record. The proposed method does not require windowing strategies in spatial direction. The shaping regularization controls the variability of nonstationary autoregression coefficients. There are two key parameters in the proposed method: filter length and radius of shaping operator. Synthetic and field data examples demonstrate that, compared with f-x deconvolution, f-x NAR can be more effective in suppressing random noise and preserving the signals.
展开▼
