Removing the Vegetation Signature from Digital Elevation Models of Coastal Areas Surveyed by Unmanned Aerial System Photogrammetry

Abstract

Unmanned Aerial System (UAS) photogrammetry is a popular method for generating digital elevation models (DEMs) of large areas in a timely and precise manner. The DEMs produced from UAS photogrammetry can be referenced to actual known elevations via groundtruthing methods using real-time kinematic global positioning systems (RTK-GPS). A common issue is vegetation can distort the DEM, creating a phantom layer above the real world elevation of the underlying substrate. The phantom vegetation layer acts as noise that must be filtered out to gain a more accurate topographical representation. The focus of this research is on barrier islands where short term sedimentation is affected greatest by storms that rapidly redistribute material and recreate new topographical features, making it paramount to know the true elevation. The research goal of this project is to apply a proven vegetation removal methodology to high quality photogrammetry derived DEMs obtained from hobbyist UAS flights in a dense coastal vegetated region. This was accomplished via extensive field campaigns along Texas Gulf Coast areas where UAS flights, groundtruthing methods, and RTK-GPS surveys were refined and systemized. Using these processes, successful flights were performed, ground control points were accurately recorded and a variety of vegetation types were analyzed through visual recognition of vegetative types, noting their locations on the model and the correlated substrate height. The result of the field campaigns was a workable high quality DEM, numerous vegetation points and accurate ground control point. With the help of multispectral sensors, which can differentiate vegetation based upon emitted wavelengths, the false elevation from vegetation was removed from the DEM. Using multivariate regression analysis, an effective error value was discovered and applied to a range of NDVI values. The resulting DEM has an uncertainty of 2 centimeters and it is expected to remove vegetative noise by as much as 75%. More accurate and fast map generation will help coastal engineers, scientists, and environmental managers to better model the complex morphodynamics of coastal systems.