Probabilistic analysis of post-failure behavior of landslides using stochastic material point method

Landslides are one of the most common natural geological hazards in the world, which result in detrimental consequences and significant damages to human life and properties every year. In many occasions, due to extensive runout of the landslides, substantial destruction can be imposed on nearby structures. Therefore, understanding post-failure behavior of landslides is significantly important as it allows for the prediction of possible catastrophic consequences and timely planning of the disaster mitigation measures. In numerical modeling analysis, reliable prediction of landslides post-failure behavior is significantly difficult due to its large deformation, inherent heterogeneity of soils, and their complex spatially varying multiple geological properties that are internally correlated with each other. The natural heterogeneous structure of soil mass makes the landslides post-failure behaviour highly random and haphazard. However, current simulation methods cannot accurately investigate the post-failure deformations. In this thesis, a new integrated probabilistic computational framework, based on the stochastic material point method, is developed to evaluate the post-failure behavior of landslides considering the effects of heterogeneous soil properties. Material point method is used to simulate the large deformations during and after the landslide failure. The spatial variability in the parameters of soils is modelled by random fields, which are discretized by Cholesky matrix decomposition method to incorporate the effects of the soil spatial heterogeneity on the post-failure deformations. A Monte-Carlo simulation is adopted to generate series of random fields with soil properties. In the computational framework, monovariate, bivariate, and multivariate geotechnical random fields are implemented to investigate the effect of different spatial varying soil properties. The capabilities of the framework in postfailure modeling are illustrated by applications to modeling heterogeneous sand collapse, v clay slope with undrained condition, and cohesive-frictional slopes. It found that the spatial varying soil properties notably influence the post-failure behavior of landslides/flows. It is demonstrated that the uncertainty of runout distance increases with the increase of coefficient of variation for soil properties. In addition, a practical landslide hazard zoning is conducted to quantitively evaluate the level of disaster for facilities or structures located in the vicinity of the slope by using the exceedance probability of influence distance and runout distance. Five categories based on different thresholds of exceedance probability are used to visualize the area potentially affected by the landslide. In comparison, the deterministic analysis particularly underestimates the post-failure behavior and the runout motions, which may give an nonconservative estimation of the potential risks for structures located in the vicinity of slopes. Moreover, the framework allows the user to consider soil interdependent anisotropy and fabric orientation in assessing the post-failure behavior of a landslide. Not only the magnitude of uncertain soil properties, but also the orientations of soil stratification are considered, which shows the proposed method is capable of reproducing the post-failure behavior of landslides with different rotational layer of heterogeneous soil. The findings highlight the importance of considering the orientation of soil stratification, rather than only the magnitude of shear strength, in analysing the runout motions of a landslide. Furthermore, based on sparse field geo-data, dependent random variables are cross-correlated using a multivariate copula function to capture their unique internal dependence structure, the use of which provides a more accurate representation of the interdependent properties of randomly-distributed soil in large-deformation modeling analysis. The results indicated that the cross-correlation and dependence structures among soil properties on the accuracy of post-failure behavior of landslides are significant and should not be ignored.