Three dimensional avalanche modelling across irregular terrain using DEM: Comparison with experiment

Abstract

Landslides are extremely destructive natural hazards that can cause significant economic damage and loss of life. Predicting the outcomes of landslides near settlements would prove useful for hazard mitigation and disaster management. Numerical simulation of landslides allows a wide range of outcomes based on different risks and mitigation strategies to be investigated at relatively low cost. For collision-dominated landslides, modelling using a continuum approach known as the Savage-Hutter method is most common. Another approach to modelling landslides is to use the discrete element method (DEM). This is a meshless numerical method which models the system at the particle level. It allows the prediction of individual collisions of particles with each other and with the surrounding terrain. DEM has previously been applied to landslide modelling using two dimensional discs (Cleary & Campbell 1993, Campbell et al. 1995, Tang et al. 2009, Thompson et al. 2009) and three dimensional spheres (Cleary & Prakash 2004). This approach neglects the important role particle shape plays in the behaviour of granular flow. The use of shaped particles in the form of super-quadrics for landslide modelling was first introduced by Cleary (2009, 2010). The non-sphericity of the particles was shown to have a strong impact on the flow, run-out distance and structure of the final deposit. In this paper, DEM simulations are compared to laboratory-scale experiments of nominally dry sand avalanches across an irregular three dimensional terrain undertaken by Iverson et al (2004). Comparisons between the simulations and experiments are used to better quantify particle shape effects and to investigate the sensitivity of the method to material properties and numerical parameters. Re-analysis of the published experimental results showed that the total volume of sand varied significantly during the avalanche, demonstrating strong dilation followed by weak compaction of the material. This has major implications for modelling since traditional Savage-Hutter models assume that the material is incompressible. Such dilation behaviour cannot be captured in this class of avalanche models. The advantage of the DEM approach is that it inherently captures such effects, which are heavily particle shape dependent. Critical information on material properties used in the experiment, such as particle shape, internal friction and cohesion, were not published in a quantitative manner suitable for use in DEM simulations. The influence of material property uncertainties on DEM avalanche predictions was investigated by comparing the volume of sand retained in a reservoir at the top of the flume. The particle angularity and internal friction were found to have the most significant effect on the retained volumes, with the volume increasing as angularity and friction increased. The particle aspect ratios had less impact on the retained volume but were still important. A small envelope of material property ranges were identified which allowed the best prediction of the retained volume. The final DEM deposit shapes within this optimal range of material properties were highly sensitive to the internal friction values. At an internal friction value of 0.46, the DEM deposit had very similar features to the experimental deposits. At an internal friction value of 0.66, resistance to the flow was too great and the deposits did not run far enough down the flume. At values of less than 0.36 the flow resistance was much lower, resulting in excessive run-out and lateral spread. These results indicate that a small increase in the internal friction of the DEM simulation from 0.46 is likely to closely match all the experimental attributes.