A fundamental assumption in seismic reflection processing, made about sedimentary basins, is that the Earth's reflectivity series is composed of randomly distributed spikes in time, space and amplitude. This produces a flat, or white, power spectrum and an autocorrelation which is a spike at lag zero. The random reflectivity assumption forms the basis for the predictive deconvolution algorithm.
In practice real reflectivity spectra show deficiency in power at low frequencies and, as a consequence, a line fitted to the power spectrum will exhibit a positive spectral slope. Previous investigations into the spectral properties of reflectivity sequences suggest this non-random response is related to local geology. In general steep slopes appear to be related to thin, repetitive, or 'cyclic', geological sequences, whilst shallow slopes are associated with thick, non-repetitive sedimentary deposits. It is possible to modify the predictive deconvolution algorithm to account for a spectral slope observed in the environment of interest. This is done by incorporating autocorrelation side-lobes, corresponding to the assumed spectral slope, in the development of the Prediction Error Filter.
This project uses the parameter of spectral slope to examine the effectiveness of a generalised or 'spectral slope' predictive deconvolution algorithm that can account for all geologically feasible spectral slopes.
Tests using purely synthetic reflectivities suggest that significant reductions in error are able to be made using the modified algorithm. In general, it was possible to reduce the error energy to less than one tenth of the error energy associated with standard predictive deconvolution. Real reflectivities analysed from the Amadeus Basin of the Northern Territory, and the Bowen and Surat Basins of Queensland returned similiar reductions in error energies to the synthetic reflectivities.
In testing the algorithm on both synthetic and real reflectivities, an important procedure was to track the deconvolution error as a function of assumed spectral slope. This mechanism also provides a practical means of determining the optimum spectral slope to be used when the algorithm is applied to real surface reflection data.