A Multidimensional Burst Noise Attenuation Algorithm for Seismic Data

Summary I propose a novel incoherent noise attenuation algorithm for multidimensional seismic data. The proposed algorithm models the signal as compressible in a suitable transform domain and the noise as additive locally stationary noise. Local stationarity means that the noise appears as stationary over small time intervals but, otherwise, the noise characteristics could be different over long time intervals and in space. This is a flexible model that covers both random noise and incoherent noise bursts clustered in time and space. Because the noise is locally stationary, its spectrum can be estimated over short time intervals using a power spectral density estimation algorithm. I show that, by using the estimated noise spectrum and a compressed sensing framework, the noise-free signal can be recovered from the measurement. The proposed method removes both random and burst noise, however, the noise attenuation can optionally be constrained to remove noise bursts only. I show the performance of the algorithm on a 3D marine data example where measurement is contaminated with periodic noise bursts in space and random burst noise spikes in time.