space. This approach seems to be robust to a reasonable
noise level in the data. In addition, only a few parameters need to be
chosen.
This technique presents some limitations that require improvements. For instance the local stepouts need to be single valued with no conflicting dips in the data. This limitation can be overcome by estimating a few dips and retaining the ones of interest only. Second, an initial 1D velocity model is needed as a starting guess. This approximation, along with the simplified geometry of the rays, prevent us from recovering lateral velocity variations for complex geology, e.g., salt environment. This limitation can be overcome by incorporating the work of Clapp and Biondi (2000).
In spite of these approximations, this new velocity estimation technique is able to recover lateral velocity variations without picking for a 2D Gulf of Mexico dataset. In addition, the updated velocity field seems to match the geological environment: we can see velocity changes across faults at various locations. We finally show that the estimated velocity perturbations yield a map of time shifts that can be used to flatten CMP gathers.
In theory, more iterations of velocity updating should be performed. One problem with more updates is the need for more sophisticated time or depth imaging algorithms. In addition, more updates would mean improving on the tomographic inversion by allowing any type of ray geometry and background slowness field. These changes go beyond the scope of this paper. We believe, however, that all these sources of improvements should be investigated further to provide a robust and picking-free interval velocity estimation tool.