Next: Conclusions Up: Liu and Li: - Previous: Shot gather

Field data test

For the field data test, we use a time-migrated dataset from Liu and Chen (2013). The input data shown in Figure 10 has 310 traces and 700 samples in the time direction with a sampling interval of 1

. Noise is mainly strong random noise caused by the surface conditions. Figure 11a and 11b are the denoised signal and the noise removed by the

deconvolution, respectively. These show that many signal events remain in the noise section. For further comparison, we apply

RNA to remove random noise (Figure 12a). The filter length of the

RNA is 12 samples and smoothing radius size is 3 samples (in frequency) $\times$ 20 samples (in space). Figure 12b indicates that while the

RNA has fewer signal events in the noise section than the

deconvolution but some weak events still remain. Figure 13a shows that the proposed

space noncausal SOPF method can produce reasonable results, in which the continuity of events and geological structure are enhanced. The filter size of the SOPF is 6 samples (in time) $\times$ 10 samples (in space) and the scale parameters are 100.0 ( $\lambda_t$ ), 1.0 ( $\lambda_x$ ), 1.0 ( $\gamma_t$ ), and 10.0 ( $\gamma_x$ ), respectively. Compared with Figure 12b, the difference (Figure 13b) between Figure 10 and Figure 13a contains no obvious events, and the computational speed of the

SOPF is faster than the iterative

RNA methods. If one chooses smaller values for $\lambda$ in equation 9, the noise component that is parallel to the signal may be partly kept, however, it reduces the damage to the signal.


data Figure 10. Field data.


fx,fxn Figure 11. Denoised result by the - deconvolution (a) and noise removed by the - deconvolution (b).


rna,rnan Figure 12. Denoised result by the - RNA (a) and noise removed by the - RNA (b).


d2,r2 Figure 13. Denoised result by the - SOPF (a) and noise removed by the - SOPF (b).

Next: Conclusions Up: Liu and Li: - Previous: Shot gather

2019-05-06