Time domain FWI in Matlab with applications to inversion of simulated VSP data

Ninoska Amundaray, Kristopher A. Innanen

The optimization function in a full waveform inversion (FWI) iterative scheme is given by the difference between observed and synthetic data. Initial difficulties are traceable to the environment and design of seismic acquisition, which are more complex in land datasets. Recent applications of FWI using vertical seismic profiles (VSP) demonstrate suitable features of these surveys to overcome some inversion challenges. In this report, we analyzed the performance of a standard FWI algorithm with simulated VSP data in Matlab, as a preparatory step to use it with real data. Inverted models converged towards the optimal solution, illustrating how the resolution is controlled by the disposition of sources and receivers in an acquisition survey. Multiscale approaches tested emphasize the usage of frequency bands to reconcile model features with the frequency spectrum of the dataset. Fast convergence of the algorithm is achieved utilizing a reverse time migration (RTM) of the residuals constructed using a crosscorrelation imaging condition, equivalent to the FWI gradient.