Estimating Time-Varying Wavelets using Deep Learning for Seismic Inversion

This paper explores the challenge of non-stationary in seismic signals for reservoir characterization in geophysics. Traditional seismic inversion methods, based on stationary assumptions, are re-evaluated with a novel deep learning approach for modelling time-varying wavelets. This technique aims to align more closely with the non-linear and complex nature of seismic data. The study leverages the F3 block dataset from the Netherlands, an open-source, diverse dataset ideal for examining non-stationary seismic data, for evaluation. The findings of this study subtly hint at an emerging focus for seismic inversion research, towards a deeper understanding of seismic wave propagation effects.