Time-variant wavelet extraction with spectral decomposition for seismic inversion

Seismic inversion assumes convolution model, which imply that seismic data is stationary. However, seismic data is non-stationary due to noise contamination and different kinds of wave propagation effects, which means that the frequency spectrum of the seismic signal could change from shallow to deep formations. Conventional seismic inversion uses a constant wavelet, which is usually generated from an average spectrum of the seismic data. This could result in the missing geological features and inaccurate rock properties estimation of the inversion results. In this paper, we propose a time-variant wavelets extraction method by using local attribute based spectral decomposition, which can decompose every seismic trace into time-frequency space. The time-variant wavelets are generated according to local frequency spectrum, which can used to construct a time-variant kernel matrix. By using this time-variant kernel matrix, we can obtain better correlation between synthetic and extracted seismograms than using constant wavelet. The method has been applied on the inversion for acoustic impedance, which also shows improved resolution and better fit to well-log measured impedance. Presentation Date: Thursday, October 20, 2016 Start Time: 9:45:00 AM Location: 155 Presentation Type: ORAL