Three-dimensional reliability analysis of unsaturated soil slope considering permeability rotated anisotropy random fields

The spatial variability is an inherent characteristic of soils. In addition to isotropic and transversely anisotropic soils, rotated anisotropic soils can also be observed in nature owing to deposition and weathering. Recent studies revealed that the spatially variable soil strength parameters can significantly affect rotated anisotropically deposited soil slope stability. However, the influence of spatial variability of saturated water permeability (ks) on this type of slope stability has been not fully understood, especially for the three-dimensional (3D) slope analysis. This paper aims to investigate unsaturated soil slope reliability with 3D ks rotated anisotropy random fields under rainfall infiltration. Due to the low efficiency of the covariance matric decomposition method (CMD) in simulating large-scale 3D rotated anisotropy random fields, a modified CMD method is proposed to save the computational cost. Moreover, comparisons between 2D and 3D random finite element analyses are also discussed. The results indicate the modified CMD method can markedly improve computational efficiency by more than 3.3 times compared with the CMD method. Only applying variations of pore water pressure (PWP) of a section or groundwater table (GWT) cannot sufficiently account for the uncertainty of slope stability. The cross-dip slope, where the dip direction of the slope is perpendicular to the dip direction of strata, has a larger reliability index (1.3−3.0 times) than the dip slope and reverse slope at the same soil bedding plane orientation. The most unfavorable soil bedding plane orientation for the cross-dip slope and the dip/reverse slope is 0° (horizontally deposited) and 30° (slightly larger than slope angle), respectively. It is attributed to the relatively low ks along these two orientations being more likely to form the aquiclude, affecting overall variations of PWP, leading to a larger standard deviation of the factor of safety and a lower reliability index. Furthermore, the reliability of cross-dip slope can only be studied in 3D, while the dip slope and reverse slope reliability can be studied with 2D analysis to decrease computational burden when the slope length does not exceed 50 m.