Uncertainty evaluation of the run-out distance of flow-like landslides considering the anisotropic scale of fluctuation in the random field of internal friction angle

Naturally, sedimented strata bring about complicated distributions for geomaterial properties with different random field parameters in different directions, e.g., the anisotropic scale of fluctuation (SOF). This study aims to reveal the variation law of landslide run-out distance under the random field of internal friction angle with anisotropic SOF by using a stochastic smoothed particle hydrodynamics (SPH) method. The Monte Carlo simulation technique, which requires the discretization of random samples and deterministic analysis of landslide flow processes, was chosen in the stochastic SPH method as the modeling method for parameter randomness. Then, the autocorrelation matrix decomposition algorithm was introduced to generate random fields with anisotropic SOF. The SPH method with the combination of the Mohr-Coulomb failure criterion and non-Newtonian fluid model was used to conduct the deterministic flow-process simulation of landslides. The deterministic SPH method is validated through the simulation of the Chengxi landslide in the Wenchuan earthquake and a simple stratigraphic model was simulated to study the influence of the anisotropic SOF on the random field of the internal friction angle. Thereafter, an illustrative landslide model was designed according to the cross-section profile of the Chengxi landslide, and the uncertainty characteristics of landslide run-out distances under different SOFs were evaluated by the stochastic SPH method. In conclusion, the influence of the vertical SOF on the landslide run-out distance is more obvious than that of the horizontal SOF, and the increase in the vertical SOF makes the fluctuation range of the landslide run-out distance larger. In addition, the distribution characteristics of landslide run-out distance under random fields of the internal friction angle with anisotropic SOF are more consistent with the lognormal distribution.