Abstract

Deep learning applications in quantifying subsurface rock properties have lately shown great potential when compared with conventional seismic inversions. Challenges exist in conventional seismic inversion techniques such as the assumption of constant phase, stationarity of wavelets, noise free data, etc., which are inconsistent with non-stationary time series field seismic signals. The conventional inversion process also lacks the ability to add to geological information that can otherwise be obtained from legacy seismic datasets.