Geological Uncertainty Quantification Using Image Warping and Bayesian Machine Learning

Quantifying the uncertainty of stratigraphic condition is an essential task in geotechnical projects. However, delineating and simulating heterogeneous stratigraphic profiles (non-stationary field), such as tectonically distorted or irregularly deposited strata from limited borehole information is still an open question and challenging task in engineering geology. In this study, a novel approach that applies the image warping technique to non-stationary field and combines it with an advanced stratigraphic stochastic simulation model is proposed to address this challenge. The image warping technique is effective to transform non-stationary field into stationary field based on the thin plate splines warping algorithm. Subsequently, an in-house developed stratigraphic stochastic simulation model can be applied to the transferred stationary field. The developed stratigraphic simulation approach integrates the Markov random field (MRF) model and the discriminant adaptive nearest neighbor-based k-harmonic mean distance (DANN-KHMD) classifier into a Bayesian framework to efficiently estimate the stratigraphic uncertainty given sparse site exploration results. To demonstrate the effectiveness of the developed approach, a synthetic case is studied using the developed approach. We envision this approach can be further promoted in industry practices for an improved risk control in geotechnical engineering.

Keywords: Non-stationary field, image warping technique, Markov random field, discriminant adaptive nearest neighbor-based k-harmonic mean distance, stratigraphic uncertainty, Bayesian machine learning.