Multiple prediction and subtraction from apparent slowness relations in 2D synthetic and field ocean-bottom cable data

Abstract

A target-oriented algorithm is developed for the prediction of multiples recorded on ocean-bottom cables by utilizing apparent slowness relations in common-source and common-receiver gathers. It is based on combining offsets and times of direct waves and primary reflections to predict multiples by matching apparent slownesses at all source and receiver locations; all higher-order multiples can be predicted by matching apparent slownesses alternately in common-source and common-receiver gathers. No knowledge of the subsurface velocity is required. Traveltimes of the direct waves and primary reflections need to be picked from common-source gathers. The subtraction of multiples involves flattening the predicted times of the multiple events, subtracting a local spatial average trace from each trace, within a fixed time window containing the wavelet of the multiple, and then shifting the data back to its original times. Tests of synthetic and field data indicate that the proposed method predicts multiples very well and removes them from seismic data efficiently with negligible affect on the primary reflections, as long as the primary and multiple reflections do not overlap in time and slowness over substantial windows in the domain in which the removal is done.

Data-driven, target-oriented, kinematic prediction and subtraction of multiples from pure and mode-converted multicomponent data

“…However, the magnitude of incident and reflected slownesses are preserved. When mode conversion does not occur at the reflection points, it reduces to the acoustic formulation previously discussed by Cao and McMechan (2010). The apparent slownesses of the incident waves measured along the topography of the free-surface, match those of the reflected waves.…”

“…This 2D elastic algorithm can be extended to 3D. See Cao and McMechan (2011) for the corresponding 3D acoustic algorithm. The reflections in 3D data are surfaces instead of lines so the apparent slowness at each point on the primary surface is a 3D vector with orthogonal horizontal components, p x and p y , in the inline and crossline directions, respectively, that both need to be matched in multiple prediction.…”

“…If amplitudes vary more rapidly across traces, a narrower window is required to ensure optimal wavelet estimation. See the description and illustrative examples in Cao and McMechan (2010). After the traveltimes and offsets of multiples are predicted from any single component, or combination of components of the elastic data, the predictions are then used to subtract multiples from all available components.…”

“…It is implemented here by computing a local wavelet by averaging the adjacent traces in a moving trace window, within a selected local time window centered on the flattened predicted traveltime trajectory of the multiple. The average local wavelet is then subtracted from the single trace at the center of each trace window, at each time sample (see Cao and McMechan [2010] for details). The multiple removal operator is reestimated at each position of the trace window.…”

Internal multiple adaptive subtraction using Huber norm

“…In the second step, usually a real seismic trace is used to construct filters that match the predicted multiple model and field data. The predicted multiple model then has the same amplitude and phase after matching filtering, thus helping to remove internal multiples (Araújo, 1994;Berkhout, 1997;Ikelle et al, 2002;Cao and McMechan, 2011). Traditional matching filters are always calculated by a residual minimized algorithm in L 2 -norm or L 1 -norm (Chapman and Barrodale, 1983;Guitton and Verschuur, 2004;Fomel, 2009;Wang et al, 2009;Ventosa et al, 2012).…”