DEVELOPMENT AND OPTIMIZATION OF A LANDSLIDE RUNOUT SCORING SYSTEM

While landslide runout distance is a key factor in assessing hazard levels and identifying dangerous areas, it is also notoriously difficult to predict. Therefore, we developed the Landslide Runout Score (LRS) system to enable the generation of predictive maps. The system relies on a database of 158 landslides of varying runout distance from three locations in northern California, Oregon, and Washington state for model development and calibration. Starting with a group of parameters identified in previous studies as potentially correlating to landslide runout, we narrow the list to three parameters that exhibit robust relationships with runout and can also be generated from commonly available maps and using GIS-based calculation. Values for planimetric curvature, upslope contributing area normalized by expected landslide area, and soil sand content are used to predict runout measured by the unitless Runout Number, defined as the length of runout divided by the square root of the landslide area. Long, medium, and short runout categories are defined with Runout Number breaks at 2.5 and 1.5, respectively. Weighting factors for the input parameters were selected to optimize the correlation between LRS and Runout Number, and the predictive capacity of the LRS was evaluated using ROC analysis. The accuracy of “short” and “long” runout predictions is 70-75%, and the accuracy including all three runout categories is 60-65%. The LRS is expected to apply to landslides in humid, temperate climates, with topography similar to that found in the Pacific Northwest of the U.S.