Improving Optical Bathymetry Estimation with Constrained Geographically Weighted Regression using Object-based Image Analysis
dc.creator | Steven Quan | |
dc.date.accessioned | 2022-01-25T19:47:15Z | |
dc.date.available | 2022-01-25T19:47:15Z | |
dc.date.issued | 2021 | |
dc.description.abstract | Bathymetry data form the underlying core information for studies in coastal and marine planning, coastal engineering, coastal hazards modeling, marine environment research, plate tectonics research and marine navigation. The shallow water coastal zone, in particular, can be challenging and / or expensive to acquire bathymetry data using traditional acoustic survey methods or airborne bathymetric LiDAR. Due to this, the development of optical bathymetry inversion algorithms which use electro-optical data to estimate water depths, have been shown to be advantageous in predicting water depths for many coastal zone applications. Optical bathymetry does have limitations, including variable quality and accuracy, and sensitivity to variable seafloor structure and water quality. This dissertation improves upon existing optical bathymetry methods by addressing a key component in model performance, variable seafloor structure and spatial heterogeneity, by delineating regions based on spatial homogeneity within the seafloor structure through object-based image analysis, and combining advanced geographically varying modeling approaches, the geographically weighted regression (GWR), to form a constrained geographically weighted regression (CGWR). Object-based image analysis (OBIA) was used to partition multispectral data imagery into multi-scale image objects based on spatial homogeneity and classify data into general seafloor structure, which were then used constrain the GWR, for empirical optical bathymetry algorithms. CGWR is shown to have increased model performance over many traditional modeling approaches, including global and regional ordinary least squares linear regression, GWR, and adaptive geographically weighted regression (AGWR) models. Moreover, the CGWR performance increases were consistent between two different empirical optical bathymetry algorithms tested. The CGWR approach was also employed using high resolution seafloor habitat classification data at different habitat and structure classification schemas, in place of the initial OBIA general seafloor structure classification. Two high resolution habitat classification schemas, geological structural component classification and benthic habitat structure classification, resulted in increased CGWR model performance over the initial CGWR approach. Further analysis on the effects of varying bandwidth size and CGWR model performance has shown an initial exponential relationship with model performance, eventually becoming asymptotic, with an increase in bandwidth size decreasing model performance. CGWR was shown to have lower magnitude standard deviation and standard error values compared to GWR and overall increased model performance at all bandwidths. Furthermore, the density of calibration data for modeling was shown to have a general positive relationship with model performance, with model accuracy decreasing with lower calibration data density. Interestingly, the results also suggest that CGWR may lose its modeling advantage over GWR at certain calibration data density levels. The results of this research present a new approach to improving optical bathymetry estimates, provides information about the scale at which spatial relationships affect regression models, and how model parameters can affect overall model performance. This works towards not only improving optical bathymetry estimates, but adds to increasing the applicability and utility of optical bathymetry algorithms themselves through accuracy improvements, as model accuracy is a limiting factor for many applications. | |
dc.identifier.uri | https://hdl.handle.net/1920/12706 | |
dc.title | Improving Optical Bathymetry Estimation with Constrained Geographically Weighted Regression using Object-based Image Analysis | |
thesis.degree.discipline | Earth Systems and Geoinformation Sciences | |
thesis.degree.grantor | George Mason University | |
thesis.degree.level | Ph.D. |
Files
Original bundle
1 - 1 of 1