Horizon Guided Initial Model for Full Waveform Inversion of Seismic Data without Low Frequencies

Abstract

Initializing the Full Waveform Inversion (FWI) iterations from sufficiently low frequencies through a multiscale scheme provides long wavelength information in the early stages of FWI and yields the convergence to the correct solution of the problem. However, because of the band-limited nature of the acquired seismic data the low frequency components are not practically observed and it is not realistic to start the FWI from frequencies lower than recording limits (<5 Hz). We propose a method for developing the low frequency component of the model using conventional data processing flows and well log data. A pre-stack time migrated (PSTM) section is used to develop a reflectivity section through sparse spikes inversion. Then a horizon guided well interpolation followed by acoustic impedance inversion of reflectivity section provides the acoustic impedance model. By using constant density model or empirical equations, the acoustic impedance is converted to velocity model. Finally, depth conversion provides a good initial model for FWI without low frequencies in the seismic data. Synthetic examples of Marmousi2 model demonstrate the efficiency of the proposed method.