Geotechnical uncertainty, modeling, and decision making

TLDR

Modeling is only one aspect of decision making, yet its practical application is limited by the lack of explicit uncertainty consideration; reliability analysis, which outputs probability of failure—a metric sensitive to data and meaningful for both system and component failures—offers a more comprehensive, statistically grounded alternative to the global factor of safety. The paper reviews how uncertainty quantification and calculation improve modeling in decision making through reliability analysis, reliability‑based design, and inverse analysis. The review examines uncertainty quantification of soil properties, stratification, and model performance to support resilience engineering’s system‑level analysis. Geotechnical software improves decision support by outputting probability of failure/reliability index alongside traditional metrics, and reveals critical non‑classical failure mechanisms arising from spatially variable soils that would be missed by conventional homogeneous or layered soil analyses.

Abstract

Modeling only constitutes one aspect of decision making. The prevailing limitation of applying modeling to practice is the absence of explicit consideration of uncertainties. This review paper covers uncertainty quantification (soil properties, stratification, and model performance) and uncertainty calculation with a focus on how it enhances the role of modeling in decision making (reliability analysis, reliability-based design, and inverse analysis). The key output from a reliability analysis is the probability of failure, where "failure" is defined as any condition that does not meet a performance criterion or a set of criteria. In contrast to the global factor of safety, the probability of failure respects both mechanics and statistics, is sensitive to data (thus opening one potential pathway to digital transformation), and it is meaningful for both system and component failures. Resilience engineering requires system level analysis. As such, geotechnical software can provide better decision support by computing the probability of failure/reliability index as one basic output in addition to stresses, strains, forces, and displacements. It is further shown that more critical non-classical failure mechanisms can emerge from spatially variable soils that can escape notice if the engineer were to restrict analysis to conventional homogeneous or layered soil profiles.