Local slope, hillslope length and upslope unstable area as 1st order controls on co-seismic landslide hazard.

Abstract:

Many communities in mountainous areas have limited access to and/or understanding of co-seismic landslide hazard maps. Furthermore these maps rarely provide the information that a community seeks: Where is safest? How big could the landslide be? Geomorphic intuition suggests that: 1) on the ridges one is less likely to be hit by a landslide than elsewhere in the landscape; 2) hazard increases with the amount of upslope unstable area; 3) longer slopes contain more candidate landslides and are also capable of producing larger landslides thus they constitute a more severe hazard. These observations could help communities in siting infrastructure or making earthquake plans but have not, to our knowledge, been tested against past landslide inventories. Co-seismic landslide models make no attempt to predict landslide size and focus on initiation, ignoring the runout which is critical in the slope length control on hazard. Here we test our intuitive hypotheses using an inventory of co-seismic landslides from the 2008 Wenchuan earthquake. The inventory is mapped from high-resolution remote imagery using an automated algorithm and manual delineation and does not distinguish between source and runout zones. Discretizing the study area into 30 m cells we define landslide hazard as the probability that a cell is within a mapped landslide polygon (p(ls)). We find that p(ls) increases rapidly with increasing slope and upslope area. Locations with low local slope (<10˚) or upslope area (<900 m²/m) have p(ls) less than one third of the areal average. The joint p(ls) conditional on local slope and upslope area identifies long steep slopes as particularly hazardous and ridges (where slope and upslope area are both low) as particularly low hazard. Examining the slope lengths associated with each landslide in the inventory we find that hillslope length sets an upper limit on landslide size but that its influence on the detailed size distribution is more difficult to untangle. Finally, we combine local slope and upslope unstable area in a simple mechanistic rule-based model of landslide runout hazard and test its ability to predict p(ls). Our findings support the intuitive view that long steep slopes are among the most hazardous locations while ridges are the least hazardous locations in terms of co-seismic landslides.