LIDAR GROUND SURFACE CLASSIFICATION IN THE MIDDLEBURY RIVER WATERSHED

The rapidly expanding coverage of high resolution LiDAR-derived topographic datasets enables many new applications in geomorphology and related fields. In the northeastern U.S., LiDAR data allows researchers to see through the forest canopy and visualize the land surface in rich detail. This study aims to develop a semi-automated method to differentiate between ground surface types in northern New England such as bedrock, fluvial/alluvial terraces and glacial till, which appear texturally distinct in high resolution digital elevation models (DEMs). Using a 12km² study area in the lower Middlebury River watershed in East Middlebury, VT, spatial statistics such as slope, roughness and variability are used to characterize the surface textures. Multiple training and classification methods were run in ArcMap and the resulting class maps were ground-truthed to assess the accuracy of the different statistic combinations and classification techniques in the study area. Standard deviation of residual topography calculated over a 5x5 moving window and residual topography maps have proven most useful for differentiating surfaces types. Although additional study is required to refine our approach, accurate classification of ground surface from LiDAR may ultimately be useful in geologic mapping, sediment budgets, and infrastructure planning.