*4.1. Synthetic Data*

In this section, we first test the proposed method on synthetic seismic data. The dataset contains a sweep signal with a 2–20 Hz frequency band to represent the ground roll and seismic wave simulated by finite difference for a three-layer model as shown in Table 1 [3,18]. According to the polarization property of ground roll, the phase of the horizontal component has a 90 degree lag compared to the vertical component. A 50 Hz Ricker wavelet was used for the body wave with a time sample interval of 1ms. The number of samples for each trace was 1001, and the number of traces for one shoot was 200. The model parameters are presented in Table 1. Figure 4 shows the ground roll attenuation results of the synthetic data obtained using the NA-MEMD method.


**Table 1.** Model parameters.

Figure 4a,d represents the horizontal and vertical components of the synthetic data, respectively. Two noise-assisted signals are embedded into the synthetic data to help overcome mode mixing when using the MEMD method. We decomposed the model data into eight IMFs. According to the NA-MEMD principle, the first two IMFs are related to noises; therefore, they are discarded. Owing to the low-frequency characteristic of ground roll, the seventh and eighth IMFs dominated by the ground roll are directly subtracted from the decomposition. We reconstructed the low-index IMFs and obtained the final data with ground roll attenuation. Figure 4b,e displays the data after removing the ground roll of the horizontal component in Figure 4a and the vertical component in Figure 4d. It can be seen that the ground roll is well suppressed. From the ground roll separated in Figure 4c,f, we can see that the residual energy of the reflection wave is weak. During the removal of ground roll using the NA-MEMD method, the number of removed IMFs is a key parameter for the suppression effect. Figure 5 shows the sensitivity for the seismic wave separation with different numbers of subtracted IMFs. We find that the more high-index IMFs are removed, the better ground roll suppression is achieved. Meanwhile, more damage occurs to the reflected wave. The parameter choice of removed IMFs is a trade-off between less damage to the reflection energy and more suppression for the ground roll.

**Figure 4.** The ground roll attenuation results via NA-MEMD: (**a**) the horizontal component; (**b**) attenuated result of panel (**a**); (**c**) ground roll separated from panel (**a**); (**d**) the vertical component; (**e**) attenuated result of panel (**d**); (**f**) ground roll separated from panel (**d**).

**Figure 5.** Comparison of the number of IMFs for the vertical component: reconstructed result using (**a**) the 3rd–7th IMFs; (**b**) The 3rd–6th IMFs; (**c**) the 3rd–5th IMFs; ground roll separated using (**d**) the 8th IMF; (**e**) the 7th and 8th IMFs; and (**f**) the 6th–8th IMFs.
