1 INTRODUCTION The mechanical properties of the rocks forming the upper part of the earth crust are affected by weath- ering, i.e. by the physical and chemical action of en- vironmental agents. The amount of bonding between rock grains and their strength decrease with time so that, at the macroscopic level, a progressive reduc- tion of rock cohesion takes place. The height of a rock cliff is directly proportional to the value of the cohesion. Its progressive reduction causes therefore slope failures and the retreat of the slope crest. Ex- amples of such a phenomenon are reported by e.g. Hutchinson (1973), Kirkby (1984), Yokota & Iwa- matsu (1999). Various types of materials are af- fected: clay shales, sandstone, chalk, pyroclastic de- posits. The speed at which slope retreat occurs is usually rather low, but, in volcanic or carbonatic rocks, it is fast enough to jeopardize the safety of buildings or infrastructures located nearby. The aim of the paper is to establish a procedure for determining such a speed as a function of weath- ering induced cohesion decrease. The results of an analytical study, based on the hypothesis that weath- ering affects uniformly the rock mass, are recalled first. Although unrealistic, such a hypothesis allows the definition of a benchmark, against which it is then possible to set up a model based on the discrete element method. By means of that, it is possible to investigate different erosion scenarios and eventu- ally take into account rock and weathering dishomo- geneities in a more realistic way. 2 STABILITY OF A VERTICAL CUT The determination of the critical height of a vertical rock cliff is one of the classical problems of soil me- chanics (Fig. 1a). The material strength is ruled by the Mohr- Coulomb failure condition (Fig. 1b): c tan τ σ φ ≤ + (1) where c is the cohesion and φ is the friction angle. a) b) Figure 1. a) Vertical cliff. b) Failure condition. For a vertical cliff, upperbound limit analysis calcu- lations give: 3.83 tan 4 crit c H π φ γ   = +     (2) where γ is the unit weight of the material. By applying finite elements and linear program- ming methods, as suggested by Lyamin & Sloan (2002), Loukidis et al. (2003) showed that lower bound solutions differ less than 3% from upper bounds, even when the slope face is not vertical. On modelling rock slope retreat by the Discrete Element Method F. Calvetti, R. Nova & S. Utili Department of Structural Engineering, Politecnico di Milano, Italy ABSTRACT: The aim of the paper is to investigate the effects of weathering on the reduction of bond strength of rocks and to determine the consequent speed of slope retreat. The results of an analytical study, based on the progressive reduction of rock cohesion with time, are recalled first. A 2-D discrete element model is set up next. A comprehensive procedure to determine the appropriate model constitutive parameters and their evolution with time is presented. The numerical predictions are compared to the results of the ana- lytical study and to the published experimental data. It is shown that reasonable agreement can be found be- tween the theoretical methods and the experimental evidence. The DEM numerical method proves however to be most efficient in taking non-homogeneities into account and in modelling different erosion scenarios.