Full-waveform inversion with randomized L1 recovery for the model updates

Full-waveform inversion (FWI) is a data fitting procedure that relies on the collection of seismic data volumes and sophisticated computing to create high-resolution results. With the advent of FWI, the improvements in acquisition and inversion have been substantial, but these improvements come at a high cost because FWI involves extremely large multi-experiment data volumes. The main obstacle is the "curse of dimensionality" exemplified by NyquistÕs sampling criterion, which puts a disproportionate strain on current acquisition and processing systems as the size and desired resolution increases. In this paper, we address the "curse of dimensionality" by randomized dimensionality reduction of the FWI problem adapted from the field of CS. We invert for model updates by replacing the Gauss-Newton linearized subproblem for subsampled FWI with a sparsity promoting formulation, and solve this formulation using the SPGl1 algorithm. We speed up the algorithm and avoid overfitting the data by solving for the linearized updates only approximately. Our approach is successful because it reduces the size of seismic data volumes without loss of information. With this reduction, we can compute a Newton-like update with the reduced data volume at the cost of roughly one gradient update for the fully sampled wavefield.

https://slim.gatech.edu/Publications/Public/Conferences/EAGE/2011/li11EA...