Abstract:

FWI（FullWaveform Inversion） which is based on the framework ofBayes theory is the leading technology of seismic exploration. The basic idea of FWI is widely used for de-noising, deconvolution, seismic data regularization, 1D impedance inversion, AVA elastic parameter inversion, pre-stack migration imaging, velocity analyzing and tomography. Due to the nonlinearity between the observed data and inverse model parameters, there are still a lot of difficulties when FWI is used for field data processing. The essence of FWI was presented under the frame of statistic theory firstly. It was pointed out that Bayes estimation can be implemented with least squares by assuming that the noise satisfies Gaussian distribution. Then the mathematics physics meaning of the mapping from data space to model space is explained. And the nonlinearity of this mapping is decided by the nonlinear relationship between data space and model space, or decided by the complexity of the model parameter and forward operator. At last, the strategy which can push the practical use of FWI was presented after the comparison of the characteristics between land data and marine data. It was believed that:①the velocity inversion should be based on CWI（characteristic waveform inversion）, but not FWI;②the proportion of phase error in error function should be increased legitimately; ③the prior information about data space and model space should be considered more adequately. Only by these ways, FWI can be pushed to its practical use.

Key words: Lohrentz curve, injection profile, nonhomogeneity, nonuniformity coefficient, uneven extent