Bayesian selection of slope hydraulic model and identification of model parameters using monitoring data and subset simulation

Accurate prediction of pore water pressure responses in soils is critical for predicting stability of a slope when subjected to rainstorms. Slope hydraulic model and soil hydraulic parameters have been widely used in prediction of pore water pressure responses under rainfall infiltration, but establishing a suitable slope hydraulic model and accurately identifying its parameters for a specific slope are often challenging due to a lack of knowledge and site investigation data on the slope, particularly the subsurface conditions. In-situ monitoring data (e.g., time series of pore water pressure under rainfall infiltration) from a slope provide valuable information about subsurface conditions of the slope. This study proposes a probabilistic back analysis method, using Bayesian updating with subset simulation (BUSS) and slope monitoring data, to determine the most suitable slope hydraulic model from a group of candidate models with different settings (i.e., governing equations, boundary conditions, and initial conditions). The most probable soil hydraulic parameters (e.g., spatially variable soil properties used in the slope hydraulic model) are identified simultaneously in the proposed method. The most suitable slope hydraulic model significantly improves quantification of uncertainties in soil hydraulic parameters and prediction of pore water pressure responses in slopes during rainfalls.

Automated summary

This study introduces a new method for determining slope hydraulic models and soil hydraulic parameters using a probabilistic back analysis approach, which identifies the most suitable hydraulic model and the most probable parameters simultaneously using slope monitoring data. This approach significantly improves the quantification of uncertainties in predicting pore water pressure responses during rainstorms.