Full waveform inversion typically defines many more variables in the inversion than can be expected to be independently recovered. This prevents usage of more powerful optimization techniques, where large numbers of variables become very expensive. Here, FWI is performed with a model resolution defined based on the data frequencies considered at each iteration. This allows for Newton optimization to be employed in the recovery of the low frequency part of the model, considerably reducing cross-talk on long wavelength scales.
View full article as PDF (0.43 Mb)