A nonlinear solution to 3D seismic data conditioning using trained dictionaries

Abstract

Seismic analysis for reservoir characterization has been a primary focus for the geophysical community for decades. One of the critical steps in delivering high-quality processed seismic data for seismic analysis is to remove undesirable prestack seismic phenomena prior to amplitude variation with offset (AVO) analysis. Contrary to the conventional approach, which is mainly 2D gather-based and assumes flat events, we have developed a 3D nonlinear approach with a single principle: the 3D geologic structure should be invariant from offset to offset. Trained dictionaries, generated by 3D complex wavelet transformation over pilot volumes, are progressively constructed by stacking over selected offsets or angles. A sparse nonlinear approximation using the L0 norm is imposed on the data against the trained dictionaries after applying a 3D complex wavelet transform to the data. The final step is to apply an inverse 3D complex wavelet transform to the sparsified coefficients to return to the data space. This workflow is repeated for all offsets or angles. The workflow is automatic and requires minimal user input, resulting in a fast and efficient process. Multiple field data examples have demonstrated significant signal-to-noise ratio uplift, AVO and azimuthal AVO conservation, preservation of steeply dipping structural events, and multiple suppression. The processing time is significantly shorter compared with alternative conventional processes.