Date of Award
Open Access Dissertation
Tidal marshes are dynamic environments, now more than ever threatened by both natural and anthropogenic forces. Best practices for monitoring tidal marshes, as well as the environmental factors that affect them, have been studied for more than 40 years. With recent technological advances in remote sensing, new capabilities for monitoring tidal marshes have emerged. One of these new opportunities and challenges is hyper-spatial resolution imagery (<10 >cm) that can be captured by small unmanned aerial systems (sUAS). Aside from enhanced visualization, structure-from-motion (SfM) technology can derive dense point clouds from overlapped sUAS images for high resolution digital elevation models (DEMs). Furthermore, Deep Learning (DL) algorithms, patterned after the brain’s neural networks, provide effective and efficient analysis of mass amounts of pixels in high-resolution images. In this dissertation, I seek to apply these developing geospatial technologies—sUAS and DL—to map, monitor, and model marsh vegetation. First, sUAS and coastal vegetation related literature was extensively reviewed to provide a secure foundation to build upon. Second, an above ground biomass (AGB) model of the tidal marsh vegetation Spartina Alterniflora was developed using high resolution sUAS imagery to assess marsh distribution and healthiness in the estuary. We determined that the best RGB-based index for mapping S. Alterniflora biomass was the Excess Green Index (ExG), and using a quadratic relationship we achieved an R2 of 0.376. Third, with a time series of sUAS missions, tidal marsh wrack was monitored before and after a hurricane event to map and monitor its short- and long-term effects of tidal wrack deposition on vegetation. sUAS proved to be an exceptionally capable tool for this study, revealing that 55% of wrack stayed within 10 m of a water body and wrack may persist for only 3-4 months over the same location after a hurricane event. Finally, deep learning remote sensing techniques were applied to county-wide NAIP aerial imagery to map Land Use/ Land Cover (LULC) changes of Beaufort County, South Carolina from 2009 to 2019, and to assess if and why marsh losses or gains may have occurred around the county from coastal development. We discovered that the DL U-net classifier performed the best (92.4% overall accuracy) and the largest changes in the county have come by way of forest loss for urban growth, which will impact the marshes over time. This dissertation advances the theoretical and application-based use of sUAS and DL to benefit application driven GIScientists and coastal managers in the coastal marsh realm to mitigate future negative impacts and expand our understanding of how we can protect such majestic environments.
Morgan, G. R.(2022). sUAS and Deep Learning for High-Resolution Monitoring of Tidal Marshes in Coastal South Carolina. (Doctoral dissertation). Retrieved from https://scholarcommons.sc.edu/etd/6777