28th January 2021
Estimating facies probabilities from seismic data
Facies probabilities can be estimated from seismic by both machine learning and probabilistic methods (Grana et al, 2020). In this talk I’ll focus on probabilistic methods, describing some of the available algorithms and comparing results. I will investigate the deeper characteristics of the problem and explain the difficulties that this presents to the standard approaches. This can inform a discussion about the challenges that would also be faced by ML methods. In particular how are prior uncertainties quantified and posterior uncertainties assessed. Uncertainty quantification always has a strong subjective element meaning that inversion results are always interpretations requiring interaction with the data.
A comparison of deep machine learning and Monte Carlo methods for facies classification from seismic data, 2020, Dario Grana, Leonardo Azevedo, and Mingliang Liu, Geophysics
Patrick Connolly is a consultant, focussed mostly on delivering training courses on AVO and inversion methods. He retired from BP in 2015 as Senior Advisor for Geophysical Analysis. His talk Probabilistic Seismic Inversion using Pseudo-Wells was awarded Best Paper at the 2017 SEG Annual Meeting.