Automated Techniques to Extract Shallow Landslide Data from LIDAR

Thursday, 18 December 2014
Adam McClure1, Jonathan D Stock2 and Jeanne M Jones1, (1)USGS, Western Geographic Science Center, Menlo Park, CA, United States, (2)US Geological Survey, Menlo Park, CA, United States
There is a small but growing catalog of storm-driven shallow landslides in areas with available LIDAR data. These datasets offer the promise of more refined testing of shallow landside models. Traditional grid-cell extraction methods may not capture slopes relevant to initiation or terminal runout for model comparison. With the intent of this study being to provide the most robust data on historic hazardous storm-driven shallow landslides, we developed a method to automate slope extraction along pathways of arbitrary length, starting at the terminal runout or headscarp and following the gradient. Using python scripting in a GIS environment, we extracted slope failure values for each landslide, thereby reducing manual analysis. Script testing was done on a 2006 landslide dataset consisting of nearly one thousand landslides we mapped from high-resolution historic imagery in Marin County, California, USA. The scripting process grouped individual landslide headscarp, lateral scarp and margins into a single entity, thereby allowing attributes to be populated for both multiple features as well as each single landslide group. Slope values were extracted above the headscarp and along the terminal runout using 10-m pathways. These values were coded for obstructions, both natural and anthropogenic, using historical imagery. This allowed us to identify both natural and man-made obstructions that could skew slope analysis results. We compared these pathway-generated values to grid cell extraction techniques to assess the degree to which the different methods capture the distributions of slopes at initiation and deposition. By developing an automated method to analyze landslide slope failure, additional historical landslide datasets with available LIDAR data can be examined rapidly, thereby contributing to the mapping of future landslide hazard zones.