CHARACTERIZING LANDFORMS AND ASSOCIATED MASS MOVEMENT PROCESSES USING AN OBJECT-BASED MAPPING APPROACH

Object-based techniques using remote sensing products for characterizing mountain landforms have the potential to improve understanding of hillslope processes and hazards. The near-term goal of this study was to develop an object-based approach for automatically mapping landforms from LiDAR topographic data. This pilot study is part of a larger modeling effort that will characterize landform susceptibility to landslides in Washington.

A principal component analysis routine was implemented for a set of environmental covariates derived from high-resolution bare-earth LiDAR topographic data. Principal components that explain >95% variability in covariate space were then segmented and classified using object-based algorithms in eCognition®, and a map of landforms was developed. The modeled landform map was then compared visually and statistically with an expert-based, manually developed map of landforms to determine the accuracy and validity of the approach. In the future, spatial densities of shallow landslides in each landform type will be computed to understand their susceptibility to shallow landslides.

The approach developed here is data-driven, integrates a wide variety of environmental covariates, and provides quantitative prediction of a wide range of multi-scale landform attributes that can be applied to slope stability analysis and hazard assessment, hydrologic models, and making decisions in land and ecosystem management.