Common practice for seismic inversion workflows in industry is to calculate individual wavelet estimates for each seismic partial angle stack, which does not take full advantage of signal redundancy. We propose a framework for simultaneous wavelet estimation from partial angle stack data by combining wavelet estimates using an empirical Bayesian formulation, after removing imaging stretch. This formalism allows one to leverage the statistical strength of additional data, while leaving open the possibility for distinct variation in wavelets on individual angle stacks. The method was tested using frequency-domain least squares wavelet estimation with multi-taper Slepian window functions on six wells from an onshore Permian 3D data set. Blind tests showed an increase in seismic-synthetic predictability when employing this method. The experiments also indicated a statistical preference for averaging wavelets across angle stacks as opposed to preserving features of the individual angle stacks, driven by the signal-tonoise ratio of the seismic data.
Presentation Date: Wednesday, October 14, 2020
Session Start Time: 1:50 PM
Presentation Time: 3:05 PM
Location: 362A
Presentation Type: Oral
