Full Waveform Tomography and Nonlinear Migration

The least-squares functional measures the difference between observed and modelled seismic data. Because it has a high computational cost and suffers from local minima, it has limited use for the inversion of model parameters. A good initial velocity model is required. Given such a model, the minimisation of the least-squares functional resembles nonlinear migration more than inversion.<br><br>Several authors observed that the model could be updated by diving waves, without the risk of ending up in a local minimum. They used frequency-domain acoustic modelling codes to construct a velocity model. This full waveform tomography is limited to a maximum depth, determined here by considering a simple model. This creates a dichotomy. Down to the maximum depth of diving waves, least-squares minimisation combines tomography and migration. Beyond that depth, nonlinear migration dominates. The dichotomy has consequences for the choice of frequencies when using a frequency-domain acoustic modelling code.<br><br>The acoustic approximation will lead to a number of problems when using long-offset data. We show that reasonable results can still be obtained on synthetic marine data created by an elastic time-domain finite-difference code. The resulting density is not correct, but the overall geometry is.<br>