Bayesian Perspective on Ground Property Variability for Geotechnical Practice

Abstract

Soils and rocks are natural heterogeneous geo-materials, and their properties exhibit site-specific spatial variability as an outcome of the previous geological processes that the soils and rocks in the site have undergone. Spatial variability of ground properties and other geotechnical uncertainties may be modelled probabilistically using random variables or random field. Some questions are frequently raised by practicing geotechnical engineers when they consider using probabilistic methods. For example, what is the physical meaning of failure probability and random variable or random field modeling? Is a large amount of data necessary for using probabilistic methods? This paper aims at providing answers to these questions from a Bayesian perspective. Bayesian methods and tools are also presented that were recently developed for characterization of ground property variability from sparse site investigation data.

Managing Risk in Geotechnical Engineering – From Data to Digitalization

“…Compressive sampling is not derived from the random field model, but originates from signal processing. Some attempts have been made to apply compressive sampling without detrending (Wang Y. et al 2019a) and without assuming stationarity (Wang Y. et al 2019b) as discussed in the preceding section. In fact, when sufficient data are available say in the form of training images, multiple point methods that consider more than two-point autocorrelation information are being explored in geostatistics (Mariethoz and Caers 2015).…”

“…On the other hand, BCS provides both the best estimate (i.e., the mean of the random field) and the covariance matrix for the signal of interest directly from sparse measurements. developed a BCS-KL random field generator to simulate RFSs directly from sparse measurements and offered a Bayesian perspective of random field modeling of site-specific spatial variability (Wang Y. et al 2019a). The BCS-KL generator is non-parametric and data-driven.…”