Frequency-domain sparse Bayesian learning inversion of AVA data for elastic parameters reflectivities

• Frequency-domain prestack sparse Bayesian learning inversion method is proposed.
• The method retrieves sparse P- and S-wave impedance reflectivity by adding, deleting or re-estimating operator.
• Parameterized Gaussian prior helps retrieve sharp layer boundaries and precondition helps improve the inversion results.
• Synthetic data and real data are adopted to demonstrate the performance of the proposed method.

The prestack amplitude variation with angle (AVA) inversion method utilising angle information to obtain the elastic parameters estimation of subsurface rock is vital to reservoir characterisation. Under the assumption of blocky layered media, an AVA inversion algorithm combining prestack spectral reflectivity inversion with sparse Bayesian learning (SBL) is presented. Prior information of the model parameters is involved in the inversion through the hierarchical Gaussian distribution where each parameter has a unique variance instead of sharing a common one. The frequency-domain prestack SBL inversion method retrieves sparse P- and S-wave impedance reflectivities by sequentially adding, deleting or re-estimating hyper-parameters without pre-setting the number of non-zero P- and S-wave reflectivity spikes. The selection of frequency components can help get rid of noise outside the selected frequency band. The precondition of the parameters helps to balance the weight of different parameters and incorporate the relationship between those parameters into the inversion process, thus improves the inversion result. Synthetic and real data examples illustrate the effectiveness of the method.