Dynamic-warping full-waveform inversion to overcome cycle skipping

In this paper, we develop Dynamic-warping Full Waveform Inversion (D-FWI) to address cycle skipping problem which is a well-known challenging issue in conventional FWI. The dynamic warping technique is used to detect the travel time difference between the predicted and the observed wave fields. We make use of the time shift to partially warp the observed data and thus generate a series of wave fields that connect the predicted and the observed wave fields. We then use these modified observed wave fields in the framework of FWI and solve a sequence of conventional FWIs to avoid cycle skipping issue. The applications on synthetic and real data examples show that D-FWI can converge successfully by overcoming the cycle skipping problem while conventional FWI result in a spurious model. With this new approach, we can get reasonable velocity model even starting with a smooth model at higher frequency. It will greatly save the processing duration since velocity model building can be done at the same time of pre-processing such as denoise, demultiple and interpolation.