I have developed a method to perform automatic velocity picks using a nonlinear Monte Carlo random search method. I solve the problem by generating random trial interval velocity models, computing their associated rms velocity paths, and integrating along the rms paths through velocity semblance scans. The interval velocity that maximizes semblance is retained as the best fit. I use various constraints to ensure that the interval velocity model is geologically reasonable.
I tested the method on 300 marine CMP gathers from the Gulf Of Mexico. The test was successful in that a reasonable 2-D rms and interval velocity model was picked that enhanced the stack and prestack migrated sections. The relative misfit error along the line averaged 20%, and can be used as a QC tool to quickly determine regions that may require manual editing and repicking of velocities, although none was required here. The Monte Carlo algorithm requires about 10 CPU seconds per semblance scan to make the optimal pick, based on run times on an IBM RS/6000 Model 530.