A GIS framework for the stochastic distributed modelling of rainfall induced shallow landslides

Deterministic distributed models to forecast shallow landslides spatially extend site-specific slope stability and infiltration models. A problem with using existing deterministic models to forecast shallow landslides is the difficulty in obtaining accurate values for the several variables that describe the material properties of the slopes, particularly over large areas. An additional problem is the operational difficulty in performing the simulations. This is because of the amount and diversity of the topographic, geological, hydrological, and rainfall data required by the numerical models. To overcome these problems, we propose a stochastic approach to the distributed modelling of shallow rainfall-induced landslides in a GIS environment. For this purpose, we developed a new stochastic version of the Transient Rainfall Infiltration and Grid-based Regional Slope-stability analysis code (TRIGRS). The new code (TRIGRS-S) uses Gaussian and uniform probability distributions to describe the mechanical and hydrological properties of the slope materials. A Monte Carlo approach is used to investigate the variability of the model parameters. To help in the preparation of the model input data, and in the execution of the simulations, we implemented the TRIGRS-S code in the GRASS GIS environment. Statistical analysis of the results is performed in R, a programming language and software environment for statistical computing and plotting. TRIGRS-S was tested in a 3-km2 area north of Seattle, USA, and in a 13-km2 area south of Perugia, Italy. Adoption of the stochastic framework in the two study areas has resulted in improved spatial forecasts of shallow landslides, when compared to the deterministic forecasts. We attribute the difference to the natural variability of the mechanical and hydrological properties of the slope materials, and to the uncertainty associated with the simplified slope- stability and infiltration models. We expect the stochastic approach, code TRIGRS-S, and its integration in the GRASS GIS environment, to improve forecasts of the spatial and temporal occurrence of rainfall-induced shallow landslides, and to aid investigating the variability of slope material properties over large areas.