The key issue for the success of 3-D pre-stack depth migration is how to make a reasonable interval velocity model in depth. In this paper, a robust method has been developed to effectively set up velocity-depth models. According to Deregowski loop, the best migration velocity can be approached by cascade migration based on the sensibility of seismic imaging to velocity variation. Within CRP gather, the concept of squared residual slowness semblance was first proposed that is similar with the routine stack velocity semblance of CDP gather in time, and it had been put into practice. The relationship between stack velocity and interval velocity was deduced in depth. On cascade migration, the initial velocity, residual velocity and modified velocity are mathematically united during the migration iterations. Within squared residual slowness semblance, the velocity picking is automatically achieved by Monte Carlo nonlinear optimizing. In order to make the picked velocity reasonable in geology, velocity constraints are given in velocity picking, and convergence criteria are forced during the Monte Carlo nonlinear optimizing. Data tests with SEG-EAGE data and 58km(2) Hailaer 3-D seismic data have shown good prospects of this method, some improved seismic imaging and accurate velocity models were achieved. This algorithm can be used to quickly establish a velocity model in depth, and it will also reduce the processing costs, that makes the method suitable for velocity analysis of large area 3-D survey.
