Bayesian analysis of soil water characteristic curve using Markov chain Monte Carlo simulation and its application on soil water infiltration

The soil water characteristic curve (SWCC) is an important property for unsaturated soils and is essential to unsaturated soil engineering analysis. There is significant uncertainty of SWCC obtained by experiment due to the complicated unmodelled influencing factors on SWCC. In this paper, regarding the fitting parameters in Fredlund and Xing (FX) model, Van Genuchten (VG) model, and Gardner model as the random vectors, the uncertainty of SWCC fitting parameters is evaluated using the Bayesian framework. This framework is demonstrated using sandy experimental data with 1,030 records in UNSODA. The posterior distributions of fitting parameters are obtained by the Markov chain Monte Carlo simulation. Different levels of confidence intervals of fitting parameters for FX, VG and Gardner models are obtained intuitively by proposed Bayesian framework. It is found that the confidence interval of the VG model is narrowest, and its uncertainty is the lowest. Different levels of confidence intervals of SWCC with VG model are applied in the one-dimensional vertical soil water filtration. The results demonstrated that the uncertainty in SWCC had significant effects on soil water infiltration.