Enabling uncertainty quantification for seismic data preprocessing using normalizing flows (NF) – An interpolation example

Abstract

Seismic data go through a sequence of pre-processing steps before being made into an image. Although some work has been done to assess the uncertainties in the final images, how the uncertainty in the pre-processing affects the results remains largely unexplored. We use Normalizing Flows (NF), a type of deep neural network, to interpolate seismic data and quantify the associated uncertainty. A big advantage of NFs, over the more commonly used Markov Chain Monte Carlo methods, is that they can successfully sample a complex and highdimensional probability space with fewer assumptions. We present the first application of NF in interpolating (synthetic) seismic data. The statistical measurements retrieved from the network can be used to better characterize the data as it is passed to the post-processing phase.