An analysis of multispectral unmanned aerial systems for salt marsh foreshore land cover classification and digital elevation model generation

Abstract

Recent advances in Unmanned Aerial Systems (UAS), and increased affordability, have proliferated their use in the scientific community. Despite these innovations, UAS attempts to map a site’s true elevation using Structure from Motion Multi-View Stereo (SFM-MVS) software are obstructed by vegetative canopies, resulting in the production of a Digital Surface Model (DSM), rather than the desired Digital Elevation Model (DEM). This project seeks to account for the varying heights of vegetation communities within the Masstown East saltmarsh, producing DEMs for mudflat/saltmarsh landscapes with an accuracy comparable to that the DSM. DEM generation has been completed in two separate stages. The first stage consists of land cover classifications using UAS derived, radiometrically corrected data. Respective land cover classifications are assessed using confusion matrices. Secondly, surveyed canopy heights and function derived heights are subtracted from their respective classes, generating the DEMs. DEM validation has been performed by comparing topographic survey point values to those modeled, using the Root Square Mean Error (RMSE) measure. The project then compares the various parameters implemented for land cover classifications, and DEM accuracy. DEM generation methods were then coupled to produce a final DEM with a RMSE of 6cm. The results suggest consumer grade Multispectral UAS can produce DEMs with accuracies comparable to the initial DSMs generated, and thus merit further studies investigating their scientific capacities.