CONCLUSIONS

The proposed method for regularizing the geometry of 3-D prestack data set performed well in a real-data test. The analysis of the data after geometry regularization demonstrates that the regularization methods fill the acquisition gaps by using the information from neighboring offsets/azimuths. Therefore, they provide a better input to the migration than the simple normalization by the partial stack fold. The imaging results confirm this analysis.

The inclusion of the AMO operator in the regularization assures better preservation of the steeply dipping event, thus yielding higher-resolution images than when the AMO operator is not applied. At the reservoir level (depth of 3.2 km), the improvements in the images are fairly subtle. We would expect more substantial differences for shallower events. However, the processing parameters (in particular, offset sampling for migration) were not optimized for shallow targets. Furthermore, in the tests presented in these paper we ignored the data azimuth. Preliminary tests indicated that taking into account the data azimuth when regularizing the geometry can improve the results at far offset. The method and the codes are ready for these further tests.

An obvious improvement to the method is to allow the the smoothing range parameter ($\rho_D$)to vary. A non-stationary regularization could better take into account the local sparseness in the data. Using $\rho_D$ that changes laterally could be dangerous and lead to instability, but making it function of depth should be trivial and makes sense because the smoothing range ought to be constant as a function of the reflection angle at depth, not of the offset at the surface.

We would like to thank Trino Salinas from Ecopetrol for bringing the data set to SEP during his visit this summer, and Ecopetrol for making the data available to SEP.