Data-driven 3D Deghosting Using Multisensor Marine Measurements

More Info


Marine seismic acquisition tows submerged streamers to record pressure waves from the subsurface. The recording, however, contains both the desired upgoing wavefield and its (immediate) reflection off the sea-surface, causing a downgoing wavefield known as the seismic ghost. Interference between the up- and downgoing waves causes periodic low signal to noise ratio (S/N) ‘ghost notches’ in the recorded spectrum. To restore the broadband upgoing signal, we must remove the ghost (‘deghosting’). Deghosting using solely pressure measurements fails to restore the signal in the low S/N notches of the data. Current acquisition techniques acquire signals with different ghost notches, such that their proper combination recovers the broadband signal. This thesis uses multisensor acquisition: measurements of the pressure and particle velocity vector. The ghost notches on the pressure and vertical particle velocity are offset by half a period, such that their combination may provide good S/N at all frequencies. Current multisensor deghosting techniques make deterministic assumptions on the data and ghost model (such as a known streamer depth, or assuming energy propagating only along the streamer). If the assumptions do not correspond to the data, the deghosting fails to restore the true broadband signal. We propose two novel data-driven deghosting techniques, which estimate an adequate deghosting filter based on the data itself. The first method estimates the 3D propagation of energy using measurements of the pressure and crossline particle velocity along a single streamer. The 3D incidence angle is used to sum the pressure wave with vertical particle velocity such that only the upgoing wave is recovered. The second method estimates the filter parameters that explain the recorded ghosted data by minimizing a multisensor least-squares deghosting cost function. The cost function is analytically shown to outperform similar single sensor adaptive deghosting techniques in terms of sensitivity to the true ghost model. The obtained filter parameters may then be used to construct an inverse filter that restores the upgoing wavefield. We found that both methods produce encouraging results on real data, outperforming the existing deterministic multisensor deghosting methods.