1D geological imaging of the subsurface from geophysical data with Bayesian Evidential Learning

Abstract

Imaging the subsurface of the Earth is of prime concern in geosciences. In this scope, geophysics offers a wide range of methods that are able to produce models of the subsurface, classically through inversion processes. Deterministic inversions lack the ability to produce satisfactory quantifications of uncertainty, whereas stochastic inversions are often computationally demanding. In this paper, a new method to interpret geophysical data is proposed in order to produce 1D imaging of the subsurface along with the uncertainty on the associated parameters. This new approach called Bayesian Evidential Learning 1D imaging (BEL1D) relies on the constitution of statistics-based relationships between simulated data and associated model parameters. The method is applied to surface nuclear magnetic resonance for both a numerical example and field data. The obtained results are compared to the solutions provided by other approaches for the interpretation of these datasets, to demonstrate the robustness of BEL1D. Although this contribution demonstrates the framework for surface nuclear magnetic resonance geophysical data, it is not restricted to this type of data but can be applied to any 1D inverse problem.