Unlocking the Low Frequencies for Improved Relative Impedance Inversion

Abstract

Abstract Seismic inversion techniques aim to estimate impedance and hence the Earth’s rock properties. Due to the band-limited nature of seismic data, estimates of absolute impedance require a low frequency scalar model estimated from horizon interpretation, well data and stacking velocities. Whatever deterministic inversion algorithm is employed, model assumptions and artefacts are embedded in an absolute impedance result. Relative impedance inversion workflows avoid the pitfalls of a low frequency model by employing the seismic data alone, and are a fast and cost effective alternative when absolute impedance is not required. A typical workflow can be summarised by: Zero phasing of the seismic data by well ties or statistical methods; Relative impedance (‘Coloured’) inversion by spectral shaping and 90 degree phase rotation; Horizon based amplitude extraction and calibration. The spectral shaping process applies a filter to the seismic reflectivity data to match it to the local well impedance spectrum. By wedge modelling we show that the range of bed thicknesses consistently imaged by the relative impedance is affected by the impedance wavelet characteristics, particularly the characteristic decay (α) of the spectrum with frequency, the low frequency cut-off, and the high frequency cut-off. The decay (α) is determined by the local fractal layering of the earth and hence the log impedance spectrum. The degree to which the high and low frequency events can be imaged by seismic is limited by noise, and we can extract maximum value from the relative impedance result when the inherent impedance wavelet has high bandwidth. Impedance data is particularly rich in low frequencies as a natural consequence of the Earth’s layering, and it is this low frequency bandwidth that new post-processing techniques aim to improve. Simple relative impedance inversions suffer from the amplification of low frequency noise if the shaping operators are pushed to their limits, creating a characteristic ‘curtain’ effect. We show how two recent processing techniques have solved this problem, extending the low frequency bandwidth often to as low as the acquisition low frequency cut-off. The first innovation is to design and apply the inversion operator before migration, which reduces the problem to a 1D convolution and reduces low frequency noise. We demonstrate the improved performance of Coloured Inversion using this technique and the added value to a data set from West Africa. The second innovation involves the application of structurally oriented filters to selected (low) frequency bands to improve signal to noise conditions prior to applying the inversion operator. We illustrate this technique with an example of relative impedance elastic inversion from the UK North Sea. Using these pre-processing techniques on seismic data it is possible to produce products with high bandwidth and correctly estimate relative impedance over a large range of bed thicknesses, delivering valuable insights to the interpreter.