Separation of Ground and Low Vegetation Signatures in LiDAR Measurements of Salt-Marsh Environments

Light detection and ranging (LiDAR) has been shown to have a great potential in the accurate characterization of forest systems; however, its application to salt-marsh environments is challenging because the characteristic short vegetation does not give rise to detectable differences between first and last LiDAR returns. Furthermore, the lack of precisely identifiable references (e.g., buildings, roads, etc.) in marsh areas makes the registration and bias correction of the LiDAR data much more difficult than in conventional urban- or forested-area applications. In this paper, we introduce reliable methods to remove random and systematic errors and to register raw data, as well as a new procedure, to determine the optimal filter window size to separate ground and canopy returns. A limited amount of field observations is used to determine the size of the filtering window which produces the minimally biased estimates of the digital terrain model (DTM). The digital surface model (DSM, representing the canopy top) is then obtained in a similar manner, and the digital vegetation model (DVM, representing the vegetation height) is computed as the difference between the DSM and the DTM. We apply this procedure to a study marsh within the Venice Lagoon, Italy, and obtain a high-accuracy DTM. The error $(z\_LiDAR-z\_field)$ is 2.2 cm, with a standard deviation of 6.4 cm. The comparison of the estimated DVM with field observations shows an underestimation of the height of the canopy top (17.7 cm, on average). The height of the lowest canopy elements (e.g., basal leaves), however, is significantly correlated to the LiDAR-derived DVM, showing that this contains useful information on the canopy structure.