Abstract
Landslide susceptibility maps based on statistical and machine learning (data-driven) methods have been successfully applied to spatially discriminate those areas more likely to initiate landslides from those less likely to be affected by slope instability. Even though such approaches enable to estimate the spatial likelihood for landslide initiation, they usually neglect the potential downslope propagation of the geomorphological process. There are however other models which can be used to assess possible runouts paths for given release areas.
This research aims to combine the outcomes of data-driven methods which are able to spatially identify potential landslide release zones, with those that allow to approximate the potential runout, informing also the downslope probabilities of an area to be further affected by landslides.
A 54km² catchment named Córrego Dantas, situated in the mountainous region of Rio de Janeiro, Brazil, is selected as the study area. After a single heavy rainfall event in 2011, 293 shallow landslides, some of which evolved into hillslope debris flows, were comprehensively mapped and were used as input observation for modelling.
First, landslide susceptibility maps based on statistical and machine learning methods are created to explore the spatial likelihood of landslides release. The best performing map is subsequently combined with the conceptual runout model r.randomwalk in order to compute the propensity of certain downslope regions to be affected. For doing that, a constrained top-down random walk approach is used. The threshold angle of reach as well as the travel distance determining the extent of the likely runout are derived by back-analysing the probability density functions derived from the observed hillslope debris flows.
The presented research contributes to a better spatial assessment of landslide-prone terrain at regional-scale by not only displaying the spatial likelihood of landslide release, but also by indicating possible downslope paths of future events.