Fast interval velocity estimation via NMO-based differential semblance

Differential semblance velocity analysis ("DSVA") flattens image gathers automatically by minimizing the mean square difference of neighboring traces in an image volume. Implementations based on normal moveout correction as "imaging" method are relatively fast, can accommodate arbitrary acquisition geometry, and can be organized to output 1D, 2D,or 3D interval velocity models. Within the limits of its imaging methodology (mild structure, data dominated by primary events), we first analyze the most stringent applicability limits which this method suffers, and describe an implementation of hyperbolic NMO-based DSVA with a number of features intended to assist in its assessment of eventual use in a production environment. Two marine 2D examples are illustrated to exhibit common features of DSVA: convergence to reasonable velocity estimates in a small number of iterations; highly aligned image gathers; agreement with standard velocity analysis and measured degradation in the presence of coherent noise. This implementation gives reasonable approximate interval velocities from data that fall within its domain of applicability at low computational cost. The results underline the importance of further research to incorporate more physics, notably multiple reflections, into the theory and practice of automatic velocity estimation.