A Detailed Examination of DTM Source Data: Light, Camera, Action

High-resolution, multi-temporal, remote sensing technologies have revolutionized geomorphic analysis. Topographic point measurements (XYZ) acquired from airborne and terrestrial laser scanning (ALS and TLS) and photogrammetry commonly are used to generate 3D digital terrain models (DTMs). Here, we compare DTMs generated using Structure-from-Motion (SfM) photogrammetry to ALS, TLS, and classic photogrammetry. Our investigation utilized 5 years of remotely sensed topographic data, from ALS (2007, 2009), TLS (2010-2012), and airborne and terrestrial close-range oblique photographs (using both classic and SfM photogrammetry) (2010-2012), of a 70,000 m², 500 m-long reach of the upper North Fork Toutle River, Washington, devastated by the cataclysmic 1980 eruption of Mount St. Helens. The study reach is sparsely vegetated and features 10-30 m-tall vertical banks separated by a 170 m-wide floodplain. In addition to remotely sensed data, we surveyed more than 300 ground control points (GCPs) using a 1-arc second reflectorless total station and map- and survey-grade GPS and RTK-GNSS. Few, if any, data sets have been obtained with this variety of technologies in spatial and temporal coincidence. We examine the application of each technique to assess fluvial morphological change, as computed by DTM differencing. A subset of GCPs was used to transform image coordinates into geodetic datum. DTM uncertainty was then quantified using the remaining GCPs. This uncertainty was used to determine the minimum level of detectable change. Owing to highly variable topography and point-to-surface interpolation techniques, method strengths and weaknesses were identified. ALS data were found to have greatest uncertainty in areas of low point density on steep slopes. TLS produced highly variable point density in the floodplain, where interpolation error is likely to be minimal. In contrast, classic and SfM photogrammetry using oblique photographs with a high degree of image overlap produced DTMs with more uniform point density. Our results indicate that each method can produce cm-level accuracy as compared to GCPs and although surface model representation varies due to point location, computed sediment volumes are comparable. Compared to classic photogrammetry, SfM uses robust image matching algorithms that appear to be better suited to oblique photographs, where scale changes rapidly with object distance. SfM-derived DTM accuracy is largely dependent on camera quality, careful photographic technique, and survey-grade GCPs. We find that SfM is an efficient and flexible solution for geomatic practitioners to create DTMs of sufficient quality to monitor geomorphic change.