Three-dimensional stochastic model for stratigraphic uncertainty quantification using Bayesian machine learning

Data-driven geotechnics is an emerging research field that contributes to the digitalization of geotechnical engineering. Among the numerous applications of digital techniques in geotechnical engineering, interpreting and simulating stratigraphic conditions with quantified uncertainty is an essential task and an open question in geotechnical practice. However, developing an uncertainty-aware integration of subjective engineering judgments (i.e., geological knowledge) and sparse objective site exploration results (i.e., borehole observations) is challenging. This investigation develops an effective three-dimensional stochastic geological modeling framework based on Markov random field (MRF) theory and Bayesian machine learning to characterize stratigraphic uncertainty. The proposed model considers both stratigraphic uncertainty (inherent) and model uncertainty (imperfect knowledge). A stratigraphic modeling example was studied to demonstrate the effectiveness of the proposed approach. We envision that this approach could be further generalized to industrial practices to improve risk control in geotechnical engineering.