AVA attribute estimation from misaligned seismic gathers using U-Net

Misalignments of primary reflections are mainly caused by inaccuracies in the velocity model used during migration. Aligning these reflections in seismic normal moveout (NMO) corrected gathers is a crucial part of seismic processing workflows prior to amplitude versus angle (AVA) interpretation. Reliable AVA analysis can only be performed after the gathers have been sufficiently flattened, however traditional approaches to event flattening typically involve utilizing parameter-rich cross-correlation-based algorithms. In this work, we have experimented with developing an approach to reliably estimate AVA attributes of events directly from non-flattened gathers. To accomplish this, we employ the U-Net architecture, trained on a synthetic catalogue comprising misaligned and aligned gathers, along with their respective intercepts and gradients of the reflectors. The efficacy of our proposed model is evaluated through testing on both synthetic and real datasets, notably including seismic angle gathers from the Volve dataset.