Monitoring Data-Driven Numerical Modeling of Slope Hydraulic Analysis Using Bayesian Updating with Structural Methods

Predicting soil hydraulic responses of a slope under rainfalls is critical for predicting slope instability. This can be performed using a physics-based slope hydraulic model and soil hydraulic parameters. However, the numerical modeling of slope hydraulic analysis may be non-trivial for some reasons. For example, subsurface conditions of the slope are invisible and may vary with time; soil properties are spatially variable; and site investigation data are often quite limited in geotechnical practice. Therefore, a slope can be modelled by a series of candidate slope hydraulic models with different choices of governing equations, boundary conditions, and initial conditions. In addition, soil hydraulic parameters from site investigation at the slope could have large uncertainties. Monitoring data (e.g., rainfall and pore water pressure of soil under rainfalls) represent actual field responses of an existing slope subjected to rainfalls and can provide valuable information for the slope subsurface conditions. This study uses monitoring data from a real slope and Bayesian updating with structural reliability methods (BUS) to select the most suitable slope hydraulic model among a series of candidate models and identify the most appropriate soil hydraulic parameters. Results showed that the most suitable slope hydraulic model not only improves quantification of uncertainties in soil hydraulic parameters, but also accurately predicts soil hydraulic responses under future rainfalls. The proposed method enables a monitoring data-driven way for numerical modeling of slope hydraulic analysis.