Adaptive ground-roll attenuation with localized eigenimage filtering

Abstract

Effective attenuation of strong ground roll while preserving the signal is still a challenging task for land seismic data processing. Eigenimage filtering, either in time domain or frequency domain, has been proposed to attenuate ground roll. The key concept is to construct a noise model from eigenimages and remove it from input data. The main task is therefore to optimize the selection of eigenimages that best construct the noise model. The amplitudes and apparent velocities of ground roll often vary with frequency, time and spatial locations. To closely model ground roll and therefore attenuate more noise without signal leakage, the selection of eigenimages needs to adapt to frequency, temporal and spatial variations. We adopt an approach that can be localized in frequency, time and space domains. The input data is first divided into several overlapping frequency bands. For each frequency band, a signal level is estimated based on reflection data outside of the ground roll cone. The eigenimage decomposition and the signal-to-noise ratio (SNR) based selection of eigenimages are conducted on localized overlapping time-space windows. In addition, to address erratic noise that is widely present in land data, amplitude outliers are excluded from the SNR estimation. Testing results show that the adaptive approach presented in this paper is robust and is able to remove most of ground roll energy while preserving the underlying signal.