6 March 2026 in SSRN Electronic Journal
This study presents a probabilistic framework for slope stability assessment that systematically accounts for stratigraphic uncertainty derived from cone penetration test (CPT) data. The framework integrates Gaussian mixture models (GMMs) for soft clustering with inverse transformation sampling (ITS) to generate ensembles of plausible soil layering configurations. This combination enables a consistent propagation of stratigraphic uncertainty into a subsequent reliability analysis. Conventional deterministic approaches and probabilistic methods rely solely on the most likely stratification. In contrast, the proposed framework quantifies the full range of plausible subsurface configurations. ITS produces multiple stratigraphic realizations that preserve the probabilistic structure of the GMM by treating each depth as a categorical distribution defined by the posterior cluster probabilities. For subsequent stability assessment, soil parameters, such as the effective friction angle and unit weight, are derived from empirical transformation models. These models take into account both observed variability and transformation uncertainties. The application of the framework to a soil slope founded on bedrock demonstrates that even moderate variations in stratigraphy can lead to substantial differences in the computed mean factor of safety (FS) and the probability of failure (Pf). The results emphasize that accounting for stratigraphic variability provides a more realistic and transparent quantification of slope reliability. The presented approach thus establishes a systematic, data-driven methodology for integrating CPT-based uncertainty into geotechnical design and risk assessment.