Incorporating spectral colour into impedance inversion

Heather Jeanne Eileen Lloyd, Gary F. Margrave

Most seismic data processing flows try to shape seismic data into a white spectrum during deconvolution. This choice of white spectra is largely a statistical convenience and most natural occurring reflectivity sequences do not have white spectra. Shaping the seismic spectra to that seen in reflection coefficients computed from well logs is known as colouring. By colouring the seismic data we are potentially rewarded with a more realistic data set. A colouring operator can be devised by dividing a modeled trace spectrum with a modeled well reflectivity spectrum. We experimented with a variety of spectral models for both data and reflectivity. We choose to model the trace spectra using a 4th order polynomial fit to the log amplitude spectrum. Alternately, we model the well reflectivity log amplitude spectrum with a 4th order polynomial log frequency. The ratio of these spectral models is our colour operator which can then be applied to the seismic data. The coloured seismic data are then input into acoustic impedance inversion using the BLIMP (BandLimited IMPedance) algorithm. The 10Hz geophone dynamite Hussar data was used in this study. When the coloured operator was applied to the seismic data, 0.2 to 0.6 seconds was overwhelmed by noisy high frequencies. These high frequencies also carried over into the impedance inversion. It seems likely that the seismic data spectrum is nonstationary and our spectral models, based on stationary spectra, are not appropriate. Further work needs to be done to perfect this method to produce quality inversions for the entire section.