Optimizing Geotechnical Site Characterization: A Value of Information Approach Using Coupled Spatial Random Fields and Bayesian Networks

Abstract

Geotechnical design is greatly influenced by spatial variability of soil properties, uncertainty of modelling, and lack site investigation information. Traditional deterministic procedures do not directly define failure probability or the economic value of additional data, while strong design optimization may produce overly conservative solutions with no explicit cost–benefit evaluation. This study proposes an integrated probabilistic program that consolidates Spatial Random Fields (SRFs), Bayesian Networks (BNs), and Value of Information (VOI) analysis to support risk-informed geotechnical site characterization. To minimize the computational burden linked with high-dimensional random fields, the spatial variability is converted into minimized variables suitable for Bayesian inference and pre-posterior decision analysis. The proposed framework allows efficient determinations of the Expected Value of Sample Information (EVSI) and Expected Net Benefit of Sampling (ENBS) for alternative investigation protocols. Two representative case studies are studied: slope stability as an ultimate limit-state problem and shallow foundation settlement as a serviceability limit-state problem. The results reveal that the optimal sampling location is mechanism-dependent, shifting from the slope toe for stability assessment to the foundation center for settlement control. Validation against independent Monte Carlo simulations shows strong agreement, with (R2 ≈ 0.95) for the reduced order prediction. The proposed framework also produces up to (143%) improvement in economic efficiency compared with conventional investigation procedures. The framework therefore presents a practical-basis for economically optimized, risk-informed site investigation planning and future geotechnical digital twin applications.