Paper Summary
Velocity model building (VMB) is among the most important problems is exploration geophysics, and it remains a challenge in many areas. An accurate, high resolution earth model is important for good quality images and accurate interpretation, particularly for reservoir characterization. We introduce a deep learning workflow that uses Fourier neural operators (FNOs) to estimate corrections to velocity models from migrated images. The workflow is akin to traditional migration velocity analysis (MVA), but it uses a neural network in place of a back projection operator. It can iteratively make high-resolution refinements to an incorrect velocity model.