Dip separation using adaptive empirical mode decomposition based dip filter

The special property of empirical mode decomposition (EMD) based dip filter (Chen and Ma, 2014) is adaptivity. In order to separate a seismic profile to obtain several dip components, the only parameter to define is the number of dip components. A brief review about EMD based dip filter is shown in Appendix A. It is worth mentioning that the total number of decomposed intrinsic mode functions IMFs ($N$) of EMD is usually limited to be less than 10 and the detectable dip components usually lay in the first 5 or 6 IMFs. The horizontal components lay in the residual for an incomplete EMD (stop the decomposition after obtaining the first several IMFs).

In the following sections, I will use the EMD based dip filter to decompose the synthetic and field data examples into 5 dip components. Note that the number of dip components is not limited to 5. As one can see, the EMD based dip filter can act as a data-driven dip components separator, without any parameters to be defined except for the separation number $N$.

Other structural filtering approaches that require a precise slope estimation will perform better when the dip estimation is applied on dip-separated profiles than the traditional implementation. The slope estimation after dip separation will be obviously superior than that without dip separation. One can use the slope estimation shown in Figure 8 as an example. The slope estimations in the separated dip components are more precise than that of the original profile in the dip conflicting area. The separated slope is smoother while the original slope is highly non-stationary when the events cross each other. The dip estimation error will transform into large filtering damages in any structural filtering approaches. Other structural filtering approaches that do not rely on the slope estimation will also benefit a lot because a reduced number of slope components will result in a much easier parameterization, such as choosing the prediction length in the prediction filtering (Chen and Ma, 2014) and choosing the minimum rank in the multichannel singular spectrum analysis (MSSA) (Oropeza and Sacchi, 2011).