Probabilistic Back Analysis For Infiltration In Unsaturated Soil Slopes

Rainfall-induced slope failures are common in many regions under tropical or subtropical climates, such as Southeast Asia, South America, Europe, part of North America and South and Southeast of China. It is important to conduct research studies about rainfall-induced landslides. In this study, probabilistic back analysis and model calibration for unsaturated soil slopes under rainfall condition are conducted based on the Bayesian theory adopting the Markov Chain Monte Carlo Simulation method.Firstly, a new probabilistic method is proposed for back analysis of slope failure. The proposed back analysis method takes advantage of both the response surface method and the Markov Chain Monte Carlo simulation method and is flexible and computationally efficient. A response surface model is employed to approximate the slope stability model and the Metropolis sampling method is adopted to calculate the posterior distributions. Two illustrative examples of back analysis of a cut slope failure and the 1997 Lai Ping Road landslide are presented. It is found that the covariance matrix of the jumping function can be set to be one half of the covariance of the prior distribution to achieve a reasonable acceptance rate. The correlation of cohesion and friction angle of soil does not affect the posterior statistics and the remediation design of the slope significantly, while the type of the prior distribution seems to have much influence on the remediation design. When only soil parameters are back analyzed, the shear strength parameters are updated more the soil permeability parameters.A parametric study is conducted to investigate the effects of saturated permeability, coefficientαof soil-water characteristic curve, water storage capacity, antecedent rainfall intensity, intensity of major rainfall, soil layer thickness and pressure head of the lower bound on the variation of pore-water pressures and advances of wetting front in a one-dimensional transient seepage model of unsaturated soils. It is found that among the three soil parameters, the saturated permeability and the coefficientαhave more significant effect on dissipation of soil suction than the water storage capacity. The antecedent rainfall, soil layer thickness and pressure head of the lower bound influence the initial pore pressure profiles significantly. However, the effects of the three boundary parameters are not significant when the duration of major rainfall is large.An illustrative example of probabilistic back analysis using the Differential Evolution Adaptive Metropolis algorithm is presented. The investigated case is a well instrumented natural terrain in Tong Chung East of Hong Kong. The field measured pore pressures are used to back analyses and calibrate the deterministic model. It is found that the posterior standard deviations of the six parameters are all greatly reduced. The coverage by the 95% total uncertainty bounds is estimated to be 0.964 for the calibration period. Compared with the result of the calibration period, the deterministic coefficients and the correlations are larger. For periods 2 to 4, the values of coverage by the 95% total uncertainty bounds are 0.52, 0.79 and 0.79, respectively. The results indicate an overall good performance by the calibrated model.