Uncrewed Aircraft Systems (UAS) can be used to collect monitoring data, including elevation information via Structure from Motion (SfM) and vegetation information from multispectral imagery, with a temporal resolution that is well-suited for dynamic barrier island environments. However, SfM data represent the elevation of the land surface including the vegetation canopy. Processing is required for estimating bare-earth elevation, which allows for understanding the underlying geomorphology.
As part of the Coastal Resource Evaluation for Management Application (CREMA) project, scientists from the USGS Wetland Aquatic Research Center and St. Petersburg Coastal and Marine Science Center produced vegetation correction models to develop bare-earth digital elevation models from SfM data for two sites on Dauphin Island, Alabama. One site was exposed to high wave energy and was dominated by beach, dunes, and supratidal herbaceous vegetation, whereas the second site was exposed to low wave energy and was dominated by intertidal marsh. To predict the elevation contribution from vegetation, the team tested several machine learning algorithms using spectral indices from UAS-based multispectral imagery and landscape position information (e.g., relative topography and distance from shore). Gaussian process regression performed the best and led to a reduction in the root mean square error for elevations of at least 29% and 69% for the high energy and low energy sites, respectively. The final error estimates were within common elevation standards for lidar-based elevation data. This effort is portable to other areas with calibration and could provide an efficient and repeatable framework for monitoring changing geomorphology and/or provide baseline elevations for predicting changes to these environments under future conditions.
Associated manuscript: https://doi.org/10.1016/j.isprsjprs.2021.08.014
Illustration of the elevation bias correction along a transect for the digital surface model (DSM) for April 2019 for Pelican Island, Alabama. Vegetated areas appear red in (a) the false-color orthomosaic (i.e., near-infrared, red, and green bands). (b) shows the corrected 0.5-meter bare-earth digital elevation model (DEM). (c) shows the relation between the corrected, bare-earth elevation with one standard deviation and the validation points.