TIME DOMAIN FULL WAVEFORM INVERSION USING A TIME-WINDOW AND HUBER FUNCTION NORM

To prevent the solution of full waveform inversion from converging to the local minimum, we present a full waveform inversion method with a data selection strategy using time-windows in the time domain. Adopting ideas from previous studies to mitigate the problem of local minima in the frequency domain full waveform inversion, we define the time-window to be associated with the highest amplitude of each observed trace. The time-window makes its corresponding data to be given greater weight in the calculation of the objective function. We apply also the Huber norm composed of a combination of l(1) and l(2) norms and use the approximated Hessian matrix both of which have been used in the frequency domain. The proposed algorithm is validated with two synthetic datasets and one real dataset. These include data from the simple anticline model, low-pass filtered data from the IFP Marmousi model, and data acquired from the Gulf of Mexico. We demonstrate that the inverted velocity models from the two synthetic datasets are in good agreement with the true models. In the real data example a reasonable velocity model is obtained which improves the reverse-time migration images.