Noise Attenuation Based on Wave Vector Characteristics

Polarization filtering has been widely used to enhance signal-to-noise ratios for multicomponent seismic data. Polarization filters routinely depend on the ellipticity and directionality of spatial particle motions. However, factors such as noise and formation heterogeneity often make the polarization characteristics of body waves hard to distinguish. Here, we introduce a technique in the time domain for the separation of valid body waves and noise based on wave vector characteristics. First, we characterise the ground-roll polarization by the median wave vectors derived in large-scale moving time windows. For the suppression of ground roll, we fit the particle trajectory of ground roll by the least square method using all components simultaneously. Second, we apply three-stage smoothing to the ground-roll-removed multicomponent records. In each stage, we use mean or median vectors derived in small-scale moving-time or moving-trace windows to attenuate random noise and other non-ground-roll related coherent noise. The filter in the proposed method is not devised according to ellipticity and directionality. Instead, we use the wave vector decomposition to distinguish between noise and valid signals. Synthetic data and field data examples confirm that the proposed method can effectively suppress noise without damaging the high and low frequencies of a valid signal