Automated Screening for AVA Classes Based on Complex-Trace Seismic Attributes

I propose a scheme for early AVA reconnaissance or screening using the near and far angle stacked seismic volumes as inputs while no suitable migrated gathers and derived AVA attributes are available. This new scheme, contrary to other screening methods, allows for an objective and explicit identification of the most used AVA classes by exploiting the unique features of each AVA class under the scope of complex-trace seismic attribute analysis. The method does not require any user parametrization and yields outcomes that preserve most of the original waveforms and amplitude dynamic range on each dedicated AVA class volume, and it also generates a single categorical volume labelled with the AVA classes, which can help accelerate the AVA reconnaissance and reduce project turn-around. The scheme benchmarks positively in the presence of mild effects resulting from some phenomena that commonly degrade the AVA integrity of data such as near-far event misalignments, signal bandwidth, angle-dependent frequency imbalances and angle-uncorrelated random noise as demonstrated on modelled data, achieving a prediction performance larger than 60-70% in terms of the F-1 score validation metric. The model and actual field applications have shown that the scheme can be implemented for use on actual seismic data to enhance early amplitude-based interpretations.