Data adaptive ground-roll attenuation via sparsity promotion

In land seismic records, low-frequency, high-amplitude ground-roll noise often obscures the seismic data and overlays important reflection information both in the t–x and f–k domains. Common processing techniques such as high-pass filtering and various f–k filtering commonly cause waveform distortions and information missing due to simple cut-off manipulation. In this study we explore the sparseness of ground-roll and reflection waveforms in the stationary wavelet transform (SWT) domain and construct a sparsity promoted signal separation scheme. Some priorly obtained signal predictions are also included in the separation scheme to model the to-be-separated signal components and enhance the separability of these two signals. We propose to obtain such predictions of the to-be-separated signal components from the separated neighboring trace by taking the coherent event assumption of seismic reflections. Since the reflection waveforms at the far offsets are less influenced by ground-roll noise, we therefore perform this trace-to-trace separation algorithm from traces at far offsets to ones at near offsets. We illustrate the method using both synthetic and field shot data. When compared with results obtained from the conventional high-pass filtering and f–k filtering methods, the results of the proposed method are better in preserving waveforms and frequency bandwidth of reflections.

► We extract ground-roll noise with a sparsity promoted signal separation method.
► Sparse signal representations are explored over the stationary wavelet transform.
► Signal predictions obtained with a data-adaptive strategy enhance the separability.
► The separation is performed trace-to-trace from far offsets to near offsets.
► Waveforms and frequency bandwidth of reflections are well preserved.