SURFACE ROUGHNESS-BASED SEMI-AUTOMATED LANDSLIDE MAPPING IN WEBER COUNTY, UTAH

Mapping landslides is a key component to developing landslide hazard maps and understanding the present risk posed by past, potentially reactivated, and future landslides. Increasing coverage of lidar elevation datasets allows for visual landslide mapping to cover an area in much greater detail than previously possible. The increased resolution slows mapping as more detail means smaller failures can be mapped. Literature exists on methods to help automate the process of landslide mapping. However, to date, no concerted effort has been made to use new lidar data and these quantitative methods to develop reconnaissance-level landslide maps.

In this project, we evaluate the efficacy of four different measures of surface roughness that have been tested in other investigations. We use these techniques to measure surface roughness of several known landslides in the Tertiary Norwood Tuff near the Snowbasin Ski Resort in northern Utah. Methods tested include the Continuous Wavelet Transform, Root Mean Squared Height, Standard Deviation of Slope, and Directional Cosine Eigenvector. From these techniques, we developed a semi-automated, computationally efficient method to calculate surface roughness. We then compared surface roughness data to landslide position data developed a priori using traditional human identification. We found that surface roughness methods can detect, in a semi-automated fashion, landslides up to ~80% effectiveness. Both the Standard Deviation of Slope and Directional Cosine Eigenvector with a window size of 25–30 m provide the highest accuracy, using 2-m-resolution lidar. However, challenges remain for the semi-automated approach. Ridges and channel bottoms are often identified falsely as landslides. Despite this, the roughness maps on their own are a useful landslide detection tool to use alongside traditional mapping methods. These maps provide a meaningful guide to a mapping geologist, highlighting potential areas of more chaotic topography, which may indicate a landslide. Ongoing work to treat lidar and elevation data as a signal to be interpreted in a numerical fashion, rather than merely used for visual analysis, continues to evolve. We believe there is a great opportunity for future work with partnering academic and applied geoscientists to bring attention to geologic hazards.