Compressive least-squares migration with source estimation

Least-squares reverse-time migration is a powerful approach for true-amplitude seismic imaging of complex geological structures. The successful application of this method is hindered by its exceedingly large computational cost and required prior knowledge of the generally unknown source wavelet. We address these problems by introducing an algorithm for low-cost sparsity-promoting least-squares migration with source estimation. We adapt a recent algorithm from sparse optimization, which allows to work with randomized subsets of shots during each iteration of least-squares migration, while still converging to an artifact-free solution. We modify the algorithm to incorporate on-the-fly source estimation through variable projection, which lets us estimate the wavelet without additional PDE solves. The algorithm is easy to implement and allows imaging at a fraction of the cost of conventional least squares reverse-time migration, requiring only around two passes trough the data, making the method feasible for large-scale imaging problems with unknown source wavelets.