Conclusions

We have developed an approach to decomposing input data into reflection, diffraction and noise components. The inverted forward modeling operator corresponds to the chain of Kirchhoff modeling, plane wave destruction (PWD) and path-summation integral migration operators. The chain of these operators accounts for the contribution of diffractions to the optimization, and guides inversion towards probable diffraction locations. Reflection model is estimated simultaneously. Separation into different components in iterative inversion is achieved by regularization: reflections are forced to be smooth along the dominant local slopes in the image domain, while diffractions are penalized by the sparsity constraints. Additional shaping regularization based on local signal-and-noise orthogonalization removes continuous events from diffractivity updates and reduces the crosstalk between the components during iterations. Incorporation of a reflection model into the inversion allows for reflected energy withdrawal from the diffraction image domain and increases the reliability of diffraction images.

Numerical experiments with synthetic data emulating high interference between reflections and diffractions demonstrates high effectiveness of the approach. A field data example confirms the robust separation of the components by the proposed approach.

The workflow has a high computational efficiency. Shaping regularization allows for fast convergence whereas the cost of the forward modeling operator at each iteration is predominated by Kirchhoff migration.