Assessment of Vertical Accuracy from UAV-LiDAR and Structure from Motion Point Clouds in Floodplain Terrain Mapping

Abstract

Remote sensing technologies are being applied to a variety of uses because of the increase in access to various products (digital sensors, UAVs, software) and its ability to model relatively large areas in a short amount of time. While these new technologies are beginning to be adopted, validation of their merit in floodplain terrain mapping is lacking. The main goal of this study is to evaluate the vertical accuracy of digital elevation models (DEMs) generated with UAV-based LiDAR and Structure from Motion (SfM), also known as photographic LiDAR or PhoDAR. Airborne (manned aircraft) LiDAR has been applied to river research in several applications and is common in many fields such as mining, archeology and surveying. SfM has been used to create digital surface models (DSMs) and DEMs of river systems and their associated riparian areas. Given the foundational difference between LiDAR and SfM technologies, the effects of vegetation on the floodplain landscape required a systematic evaluation to determine which techniques are appropriate for floodplain terrain mapping. We collected remotely sensed data from UAV LiDAR and SfM methods at four field sites located within the interior Columbia Basin and analyzed the resulting point clouds and DEMs to determine their vertical accuracy by comparing their elevations with ground surveyed data (checkpoints) throughout the study areas. Both LiDAR and SfM point clouds were filtered and the remaining ground points were compared to surveyed elevations both in their raw form and interpolated DEMs to assess their vertical accuracy. The results show that in vegetated ground cover, LiDAR point clouds were able to produce higher accuracy returns and a resulting higher accuracy DEM than SfM results. In non-vegetated areas, the accuracies between SfM and LiDAR returns are closer but still show higher accuracy from LiDAR point clouds. Ground filtering is shown to be a limitation on DEM vertical accuracy because of the inclusion of non-ground points in the filtering process. This limitation impacts the vertical accuracy of both LiDAR and SfM interpolated DEMs in floodplain habitats.