Graduate Thesis Or Dissertation
 

Advancing Spatial Data Engineering and Analysis: Integrative Approaches with GIS, Statistical Modelling, and Deep Learning

Public Deposited

Downloadable Content

Download PDF
https://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/8w32rf20m

Descriptions

Attribute NameValues
Creator
Abstract
  • Geospatial data analysis is a multifaceted discipline encompassing the collection, processing, and visualization of diverse datasets. It models and delineates the interactions of people, objects, and phenomena within geographical spaces and predicts patterns based on the relationships between different locations. Despite significant advancements in geospatial data engineering and analytics over recent decades, several challenges persist across multiple geoscience applications. Issues such as data integration and interoperability, the quality and accuracy of data alongside inherent uncertainties, managing big data, the complexity of spatial analysis, enhancing user experience and accessibility, and the modeling of three-dimensional and temporal data all represent ongoing obstacles. In essence, although current algorithms facilitate the extraction of precise geoinformation, substantial scope remains to refine and advance geospatial data processing. This dissertation develops advanced geospatial analysis techniques for coastal and water surface applications, leading to a series of research inquiries that are summarized below. The first manuscript explores the application of GIS editing, deep learning, and additional programming techniques to delineate estuarine fronts with a balance of speed and accuracy. Remote sensing offers significant advantages in frontal identification as compared to in-situ measurements, which are accurate but limited and costly. Remote sensing methods (e.g., optical sensing, infrared, and imaging radar) are indirect measurements but have the advantage of expansive spatial coverage and are more cost-effective. However, manual identification of fronts is time consuming and handling large datasets remain challenging. Herein, a novel method for delineating oceanic fronts from UAV-gathered optical remote sensing data using a deep learning approach is presented. The highly-resolved frontal data set is then used to estimate kinematic properties of the fronts. The second manuscript analyzes coastal data to assess and rank the vulnerability of precarious sites along Highway 101. To facilitate complex data processing in the vulnerability assessment, GIS, statistical methods, and Python scripting were utilized to calculate a Coastal Vulnerability Index (CVI) to assess and prioritize mitigation efforts for infrastructure in coastal areas. While the foundational structure of the CVI model employed is similar to those in previous studies, our method is novel in its incorporation of uncertainty estimates and its adaptation to factors particularly relevant to erosion on the Oregon Coast. Additionally, we apply diverse weighting factors to analyze the hierarchies of the variables. Furthermore, this study conducts a thorough examination of the significance of the parameters to identify possible simplifications of the model that other CVI studies do not consider. Based on the data from this research, an online interactive map was developed to assist stakeholders with decision making processes associated with mitigation of erosion hotspots on Highway 101. The third manuscript delves into the application of lidar data processing techniques in GIS to forecast seacliff erosion rates and trends along the Oregon coast. Estimation of historical erosion rates is difficult given the high spatial and temporal nature of erosion and relatively limited data available to capture these trends. These challenges are exacerbated when attempting to forecast erosional patterns to identify infrastructure that will be at risk in the future. This study introduces an innovative spatio-temporal algorithm designed to extrapolate erosion progression based on time series terrestrial lidar data to forecast the progression of coastal erosion. Two Digital Elevation Model (DEM) interpolation methods are explored (Triangulated Irregular Networks (TIN) transformation and Empirical Bayesian Kriging (EBK) interpolation). Additionally, the extrapolation capabilities of multi-dimensional analysis methods such as linear regression, polynomial fitting, and harmonic trend analysis are explored. The research considers several scenarios contrasting data utilized for prediction versus data used for control, ranging from 6 to 11 epochs for prediction and 6 to 1 epochs for control, respectively. The findings identify the optimal mix of settings, number of epochs, and other parameters to accurately forecast future erosion progression using multidimensional time series analysis. Ultimately, the interpolation and multi-dimensional analysis results all provided similar levels of accuracy in forecasting apart from the polynomial fitting, which showed poor performance. Additionally, in some epochs, the harmonic trend resulted in some localized extreme values. Combined, these manuscripts provide a significant advance in geospatial big data analysis by employing technologies such as spatial statistics and deep learning to inform diverse coastal applications.
Contributor
License
Resource Type
Date Issued
Degree Level
Degree Name
Degree Field
Degree Grantor
Commencement Year
Advisor
Committee Member
Academic Affiliation
Subject
Rights Statement
Publisher
Peer Reviewed
Language

Relationships

Parents:

This work has no parents.

In Collection:

Items

Thumbnail Title Date Uploaded Visibility Actions
file details KrivovaMaria2023.pdf 2024-01-09 Public Download