Towards Posterior Mapping for Interpretable Geology-Geophysics Data Fusion

Abstract:

Quantification of uncertainty in geological models is crucial for inference over geological features and histories to inform management of business risk. The unification of three-dimensional geological models and geophysical sensor models with techniques of Bayesian inference presents a principled framework for uncertainty quantification, made possible by a new generation of modular software tools. We present a series of experiments associated with navigating the posterior distribution of parameters for a simple kinematic geological model that includes anti-aliasing to avoid artefacts from model discretization, mapping the conditional dependence structure of geological parameters, and efficient Markov chain Monte Carlo proposals for posterior sampling based on information geometry. These elements can all be readily incorporated into more sophisticated future modeling workflows for implicit and/or kinematic geological models.