Weighted stacking of seismic AVO data using hybrid AB semblance and local similarity

Synthetic examples

The proposed method of hybrid AB semblance velocity analysis and local-similarity-weighted scheme, taking the near-offset trace as the reference trace, is applied to two synthetic AVO CMP gathers from ray-based modeling, two field CMP gathers, and a 2D prestack dataset from the Gulf of Mexico. All data are equipped with AVO phenomena, especially synthetic data are added with an obvious polarity-reversal anomalies.

In Figures 3a and 3b, one synthetic AVO CMP gather (noise-free and SNR=7.338) with four reflection events. Polarity reversals exist in the third and fourth events. Figure 3c and Figure 3d show conventional semblance and AB semblance using the data in Figure 3b. The third and fourth reflection events are energy smeared in the conventional semblance but focused in the AB semblance. The appearance of AVO II anomaly in conventional semblance reduces the accuracy of velocity analysis shown by the erroneous picking in Figure 3c. Figure 4 represents the NMO correction of the CMP gather in Figure 3b using the picked velocities from the conventional semblance and the AB semblance. Because of good data quality and small offset, the NMO-corrected CMP gathers look similar even though there is difference in the accuracy of the velocity analysis. Figure 5 compares the different processing methods: conventional semblance and conventional equal-weight stacking in Figure 5a, conventional semblance and SNR-weighted stacking by Neelamani et al. (2006) in Figure 5b, and AB semblance and local-similarity-weighted stacking in Figure 5c. It is found that both Figures 5a and 5b bear strong random noise in the stacked trace. The third and fourth events are stacked into the tuning artifacts of the thin interbeds (Hamlyn, 2014). The false deep interbeds arise from inaccurate velocity analysis in the conventional semblance, and the use of inappropriate weighting functions in the conventional stacking method and the SNR-weighted stacking method. The proposed approach produces the best stacked trace regarding the highest SNR and correctly detecting all true locations of reflections. In this case, the lower-amplitude stacked result for the fourth event by our proposed method needs further investigation.

hcomb-1
Figure 3. (a) Synthetic AVO CMP gather (noise-free); (b) Synthetic AVO CMP gather (SNR=7.338); (c) Conventional semblance; (d) AB semblance.

hcomb-2
Figure 4. NMO correction of the CMP gather in Figure 3b. (a) NMO-corrected gather using the picked velocity from the conventional semblance; (b) NMO-corrected gather using the picked velocity from the AB semblance.

hcomb-3
Figure 5. Stacked trace of the CMP gather in Figure 3b. (a) Stack by conventional semblance and conventional stacking; (b) Stack by conventional semblance and SNR-weighted stacking; (c) Stack by AB semblance and local-similarity-weighted stacking.

Figures 6a and 6b show another synthetic AVO CMP gather (noise-free and SNR=-3.449) with four reflection events. All reflection events are equipped with polarity reversals. Figures 6c and 6d show conventional semblance and AB semblance using the data in Figure 6b. All reflection events are energy smeared in the conventional semblance but focused in the AB semblance; however, there still exist small areas of energy focusing in the conventional semblance that indicates certain random noise contributes to the performance of semblance. The existence of AVO II anomaly reduces the accuracy of velocity analysis as shown by erroneous picking in Figure 6c. Figure 7 represents the NMO correction of the CMP gather in Figure 6b using the picked velocities from the conventional semblance and the AB semblance. The NMO-corrected gather from the conventional semblance shows some NMO over-correction due to the lower picked velocity. Figure 8 compares the different stacking methods: conventional semblance and conventional equal-weight stacking in Figure 8a, conventional semblance and SNR-weighted stacking in Figure 8b, and AB semblance and local-similarity-weighted stacking in Figure 8c. The SNR from low to high is Figure 8a, Figure 8b, and Figure 8c. In Figures 8a and 8b, shallow (0-0.25 s) strong artifacts exist which are removed by our approach in Figure 8c, and false interbeds appear clearly in all four reflections. These artifacts are caused by inaccurate velocity analysis and low SNR in the prestack CMP gather. Our approach produces the best stacked trace with the highest SNR and correctly detects all the true reflection locations. From the two synthetic tests, it could also be concluded that my proposed approach is suitable for stacking on both high- and low-SNR data.

lcomb-1
Figure 6. (a) Synthetic AVO CMP gather (noise-free); (b) Synthetic AVO CMP gather (SNR=-3.449); (c) Conventional semblance; (d) AB semblance.

lcomb-2
Figure 7. NMO correction of the CMP gather in Figure 6b. (a) NMO-corrected gather using the picked velocity from the conventional semblance; (b) NMO-corrected gather using the picked velocity from the AB semblance.

lcomb-3
Figure 8. Stacked trace of the CMP gather in Figure 6b. (a) Stack by conventional semblance and conventional stacking; (b) Stack by AB semblance and SNR-weighted stacking; (c) Stack by AB semblance and local-similarity-weighted stacking.

Weighted stacking of seismic AVO data using hybrid AB semblance and local similarity