The multiple filtering results are displayed in figure . These results were obtained by simply windowing the zero slowness slice of each inversion result displayed in figures , + , and then forward modeling. The windowing of the primary events in the model2 case involved three slowness slices due to the negative dips present on the primaries. Like the primaries, the multiples will have a component of very nearly zero slowness at the nearest offset. This energy will map into the zero slowness panel with the primaries. This energy appears as stubs at the near offsets of the results in figure . The reflecting boundary conditions contribute to the amplitude of these stubs.
I have prepared results of the PRT multiple suppression method as a comparison. The estimated primaries of the NMO-corrected data using the PRT multiple suppression method are displayed in figure . The PRT transform of the NMO-corrected data are displayed in figure . Multiple energy is obviously present in all three results in figure , the most notable cases are the distinct strong multiple ghosts present with the estimated primaries of model1 and model2. These ghosts are caused by the inability of the PRT to compactly represent the significantly non-parabolic multiples in PRT transform space such that the masking procedure can not separate the primary from the multiple energy. Examination of figure reveals that multiple energy spreads over a large region of the PRT transform space and is probably not properly isolated by the masking procedure associated with the PRT method.
The difference between the estimated primaries from each method and the original multiple-free primaries appear in figure for the beam stack method and in figure for the PRT method. The PRT method leaves a significant amount of multiple energy in the multiple suppressed results for all three cases at all times and offsets, where as the beam stack method contains multiple energy only at the near offsets. Increasing the sampling of the slowness parameter or range of the slowness parameters of the PRT transform does not change the PRT result significantly.