Modeling and subtraction of coherent noises (multiples)
According to our experience, at the first stage of cogerent noises elimination two-step algorithms are usefull (1 - prediction, 2 - adaptive subtraction). To get better results you can also combine different algorithms, for example to remove residual noises, one can apply kinematical filters.
Prediction and elimination methods of coherent noises in Prime
Based on the kinematical features of waves and noises:
- stacking, fan-filtering, linear/non-linear Radon transform, non-stationary deconvolution;
- Methods, based on the wave field transformation aimed at multiple prediction. Prediction is made as by current depth-velocity model usage, as without in, for example SRMP. Obtained model is adopted to initial wave field and then adaptive subtraction is applied;
- Filter-mask: searching and subtraction of multiples by kinematical criterion, but for operator construction model prediction is needed.
Why adaptive subtraction is so actual and needed?
Noise model, obtained by any of methods is always differ from real noise waves. It is impossible to obtain the correct amplitudes when you only shift your wave field to another time, which corresponded to the expected time of multiple wave, computed from time on primary event. (It corresponds to the standard predictive deconvolution method). The reason of adaptation on subtraction stage is caused by need of non-stationarity taking into account during filter construction. Non-stationarity is explained by following: geometrical spreading, frequency-dependent attenuation, presence of coherent noises with individual frequency range (different from the range of useful signal).
