Accurate characterization of marsh elevation and landcover evolution is important for coastal management and conservation. With recent advances in the application of high-resolution geodetic remote sensing techniques, accurate and rapid acquisition of topographic and geomorphic data is possible at a fine spatial resolution over large spatial extent. These techniques include airborne-based methods, such as Airborne Laser Scanning (ALS) and Unmanned Aircraft System (UAS) photogrammetry using Structure-from-Motion (SfM), and ground-based methods, such as Terrestrial Laser Scanning (TLS). Products such as Digital Elevation Models (DEMs), above ground biomass, and fractional vegetation cover produced by these high-resolution measurement techniques are revolutionizing and accelerating our understanding of geomorphological processes and landform transformations.
Hyperspatial three-dimensional (3D) point cloud data generated by TLS and UAS-SfM can provide sampling of the marsh scene at previously unforeseen spatial detail. However, these data have different characteristics and different representations of the underlying terrain and landcover. While there are differences, the challenges associated with these technologies are large data sets, often larger than 100 million points for a study area. Given their size and complexity, terrain mapping and extraction of relevant information from the complex 3D point cloud cannot be done without the use of intelligent algorithms. This research develops advanced geodetic imaging and computational techniques to better resolve spatial patterns in marsh elevation and landcover over larger spatial extents than is feasible using standard field surveying methods. This includes:
(1) Development of a novel unsupervised learning method for robust segmentation of TLS point cloud data acquired in marshes into homogeneous features (clusters) to reduce scene complexity for measurement and monitoring of surface and vegetation evolution. (2) Assessment and evaluation of the transferability and adaptability of the unsupervised learning method developed for a 3D point cloud derived from TLS data to a 3D point cloud derived from UAS-SfM. (3) Development of ensemble neural networks for modelling DEM uncertainty. These networks are able to estimate DEM error, and then apply the DEM correction to every raster cell while providing an uncertainty for every correction. Furthermore, these networks can be applied to any 3D point cloud with a flexibility in the number and selection of model inputs, any type of geomatics measurements, and applicable to any type of environments.
This work combines TLS and UAS-SfM measurements of a coastal marsh with the development and application of machine learning. The novel machine learning methods are developed to segment multi-perspective 3D point clouds and to correct DEMs while quantifying the spatial variability of the tolerance interval of the predictions (DEM uncertainty). The resulting information allows for enhanced change detection analysis of short-term marsh surface evolution in varying regimes. The developed computational techniques are generalizable to a wide range of coastal problems beyond marsh observations that rely on 3D point cloud data from a variety of scanning and imaging modalities, whether derived from lidar, UAS photogrammetry, or other survey methods. The resultant methods can be applied to estimate change detection uncertainty and monitor marsh evolution with these surveying technologies. In return, this can help support regional management of marshes in terms of understanding their short-term dynamics and long-term resilience.
|Advisor:||Starek, Micheal J.|
|Commitee:||Gibeaut, James, Moreno, Miguel, Prouty, Daniel, Tissot, Philippe E.|
|School:||Texas A&M University - Corpus Christi|
|Department:||Coastal and Marine System Science Program|
|School Location:||United States -- Texas|
|Source:||DAI-B 80/11(E), Dissertation Abstracts International|
|Subjects:||Geographic information science, Geophysical engineering|
|Keywords:||DEM uncertainty, Ensembles neural networks, Marsh, Segmentation, TLS, UAS|
Copyright in each Dissertation and Thesis is retained by the author. All Rights Reserved
The supplemental file or files you are about to download were provided to ProQuest by the author as part of a
dissertation or thesis. The supplemental files are provided "AS IS" without warranty. ProQuest is not responsible for the
content, format or impact on the supplemental file(s) on our system. in some cases, the file type may be unknown or
may be a .exe file. We recommend caution as you open such files.
Copyright of the original materials contained in the supplemental file is retained by the author and your access to the
supplemental files is subject to the ProQuest Terms and Conditions of use.
Depending on the size of the file(s) you are downloading, the system may take some time to download them. Please be