Shallow landslides cause damage in mountainous regions on an annual basis. The occurrence of landslides is related to heavy rainfalls or snowmelt but no unique relationship between rainfall patterns and landslides were found that could be used for prediction. The difficulties related to landslide predictions are mainly caused by the following reasons:i) the relationship between landslide frequency and magnitude is a power-law without characteristic size of landslidesii) hillslopes are heterogeneous in terms of topography, vegetation patterns, soil and hydro-mechanical propertiesiii) the triggering mechanisms that cause the sudden release of landslides are poorly understood In this project we will develop and test new models for triggering mechanisms taking into account the heterogeneity of the slope. In classical models to predict landslides, stability calculations are based on hillslope properties varying continuously in space and failure is described as continuous deformation. However, real hillslopes are heterogeneous consisting of discrete elements like tree roots, boulders and soil layers with properties changing abruptly in space. These elements may break releasing load to the neighbored elements initiating a chain reaction. In our model approach we describe landslides as a consequence of such chain reaction. This approach is inspired by concepts denoted as Self-Organized Criticality and Fiber Bundle models that describe the breaking and load redistribution in composite materials. Similar to statistics of landslides, both types of models show a power-law relationship between frequency and magnitude of elements breaking during loading.To apply such model concepts for predicting landslide patterns, the spatial distribution of the elements and the loads must be known. Vegetation patterns and surface properties can be deduced from aerial photos and digital elevation maps, subsurface structures can be revealed by geophysical measurements. The spatial distribution of the water content is a key factor determining the load and strength of the stabilizing elements and must be modeled accordingly. To test the model approach, the measured occurrence of landslides must be compared to predictions. For that purpose we will use six landslide inventories of the Swiss federal institute for forest, snow and landscape research WSL. Three phases of the study can be distinguished:i) Firstly, the spatial patterns of landslides in the inventories are compared to spatial distributions of vegetation and characteristics of the topography.ii) In a second phase the water distributions in the slopes of the six inventories will be computed and compared to the spatial pattern of landslides.iii) The last and most important part of the project is the inclusion of failure mechanism in the hydrological model with the explicit simulation of chain reactions that may result in landslide triggering.