Probabilistic Approaches for Geotechnical Site Characterization and Slope Stability Analysis

Abstract

In the last few decades, reliability-based design (RBD) approaches/codes and probabilistic analysis methods, such as probabilistic slope stability analysis with Monte Carlo Simulation (MCS), have been developed for geotechnical structures to deal rationally with various uncertainties (e.g., inherent spatial variability of soils and uncertainties arising during geotechnical site characterization) in geotechnical engineering. Applications of the RBD approaches/codes and probabilistic analysis methods in turn call for the needs of probabilistic site characterization, which describes probabilistically soil properties and underground stratigraphy based on both prior knowledge (i.e., site information available prior to the project) and project-specific test results. How to combine systematically prior knowledge and project-specific test results in a probabilistic manner, however, is a challenging task. This problem is further complicated by the inherent spatial variability of soils, uncertainties arising during site characterization and the fact that geotechnical site characterization generally only provides a limited number of project-specific test data. This study aims to address these challenges in probabilistic site characterization. A Bayesian framework is first developed for geotechnical site characterization, which integrates systematically prior knowledge and project-specific test results to characterize probabilistically soil properties and underground stratigraphy. The Bayesian framework addresses explicitly the inherent spatial variability of soils and accounts rationally for uncertainties arising during site characterization. It is general and equally applicable for different types of prior knowledge and different amounts of project-specific test data. When the project-specific tests (e.g., standard penetration tests) only provide sparse data, the Bayesian framework is integrated with Markov Chain Monte Carlo Simulation (MCMCS) to develop an equivalent sample approach that generates a large number of equivalent samples