Regional Assessment of Storm-triggered Shall Landslide Risks using the SLIDE (SLope-Infiltration-Distributed Equilibrium) Model

The key to advancing the predictability of rainfall-triggered landslides is to use physically based slope-stability models that simulate the dynamical response of the subsurface moisture to spatiotemporal variability of rainfall in complex terrains. An early warning system applying such physical models has been developed to predict rainfall-induced shallow landslides over Java Island in Indonesia and Honduras. The prototyped early warning system integrates three major components: (1) a susceptibility mapping or hotspot identification component based on a land surface geospatial database (topographical information, maps of soil properties, and local landslide inventory etc.); (2) a satellite-based precipitation monitoring system (http://trmm.gsfc.nasa.gov) and a precipitation forecasting model (i.e. Weather Research Forecast); and (3) a physically-based, rainfall-induced landslide prediction model SLIDE (SLope-Infiltration-Distributed Equilibrium). The system utilizes the modified physical model to calculate a Factor of Safety (FS) that accounts for the contribution of rainfall infiltration and partial saturation to the shear strength of the soil in topographically complex terrains. The system's prediction performance has been evaluated using a local landslide inventory. In Java Island, Indonesia, evaluation of SLIDE modeling results by local news reports shows that the system successfully predicted landslides in correspondence to the time of occurrence of the real landslide events. Further study of SLIDE is implemented in Honduras where Hurricane Mitch triggered widespread landslides in 1998. Results shows within the approximately 1,200 square kilometers study areas, the values of hit rates reached as high as 78% and 75%, while the error indices were 35% and 49%. Despite positive model performance, the SLIDE model is limited in the early warning system by several assumptions including, using general parameter calibration rather than in situ tests and neglecting geologic information. Advantages and limitations of this model will be discussed with respect to future applications of landslide assessment and prediction over large scales. In conclusion, integration of spatially distributed remote sensing precipitation products and in-situ datasets and physical models in this prototype system enable us to further develop a regional early warning tool in the future for forecasting storm-induced landslides.