Least-squares joint imaging of primaries and multiples

Multiple reflections provide redundant, and sometimes additional, information about the corresponding primary reflections. I implement a least-squares inversion scheme to jointly image (by normal moveout) primaries and multiples, with the goal of enforcing consistency between the images and the input data. Furthermore, to effect noise (“crosstalk”) suppression, I introduce a novel form of model regularization which exploits kinematic similarities between imaged primaries and multiples, and which also preserves the amplitude-versus-offset (AVO) response of the data. In tests on synthetic data, my approach exhibits good noise suppression and signal preservation characteristics. Real data tests highlight the need for careful data preprocessing. Future work points toward use of migration as the imaging operators, to exploit cases where multiples actually exhibit better angular coverage than primaries, and thus add new information to the inversion.