Transforming physically-modelled datasets for analysis via a modified FWI procedure

Joe Wong, Kevin L. Bertram

The goal of FWI (full waveform inversion) is to iteratively modify a smoothed digitized 2D velocity field so as to obtain one yielding synthetic seismograms that closely matches observed seismograms. Synthetic seismograms are calculated using a 2D velocity field that may have tens or even hundreds of thousands of pixel values that are treated as unknown variables. At each iteration, a correction to each and every one of the digital velocity values must be found to slowly reduce the differences between the calculated and observed seismograms. The large number of unknown variables make non-uniqueness due to many local minima in the misfit function a high possibility. Standard FWI procedures operating on physically-modelled seismic data have been found to have difficulty converging to acceptable answers. The problem arises from the fact that seismograms acquired using physical models have frequencies that are too high for normal procedures. This document reports on how the observed seismograms can be transformed to have lower dominant frequencies and on how a modified FWI procedure may be used to invert physically-modelled seismograms.