Applications of Random Dynamic Shot Decimation in Full Waveform Inversion

Full waveform inversion retrieves the earth model parameters by minimizing the difference between observed and modelled data. This is an iterative process with a computational cost proportional to the number of wave equation modeling steps we carry out. To reduce the cost, we can decimate the number of shots. In a fixed decimation setup, we discard part of the data that have been recorded. To harness the most of the information content in the recorded data, we can instead use a dynamic shot decimation, changing the shots we include every few iterations. We discuss the applications of random dynamic decimation and its possible advantages compared to static shot decimation. This approach works with any type of acquisition. However, the factor we can gain depends on the full waveform approach, on the types of waves (reflection versus transmission) we invert and on the acquisition geometry. We illustrate the discussion with a 2D synthetic short-offset marine example and with a 3D wide-azimuth, long-offset, low-frequency land data set.