Semester of Graduation

2020 Spring

Degree

Master of Science (MS)

Department

Geography and Anthropology

Document Type

Thesis

Abstract

Urban land use and land cover (LULC) mapping has been one of the major applications in remote sensing of the urban environment. Land cover refers to the biophysical materials at the surface of the earth (i.e. grass, trees, soils, concrete, water), while land use indicates the socio-economic function of the land (i.e., residential, industrial, commercial land uses). This study addresses the technical issue of how to computationally infer urban land use types based on the urban land cover structures from remote sensing data. In this research, a multispectral aerial image and high-resolution LiDAR topographic data have been integrated to investigate the urban land cover and land use in New Orleans, Louisiana. First, the LiDAR data are used to solve the problems associated with solar shadows of trees and buildings, building lean and occlusions in the multispectral aerial image. A two-stage rule-based classification approach has been developed, and the urban land cover of New Orleans has been classified into six categories: water, grass, trees, imperious ground, elevated bridges, and buildings with an overall classification accuracy of 94.2%, significantly higher than that of traditional per-pixel based classification method. The buildings are further classified into regular low-rising, multi-story, mid-rise, high-rise, and skyscrapers in terms of the height. Second, the land cover composition and structure in New Orleans have been quantitatively analyzed for the first time in terms of urban planning districts, and the information and knowledge about the characteristics of urban land cover components and structure for different types of land use functions have been discovered. Third, a graph-theoretic data model, known as relational attribute neighborhood graph (RANG), is adopted to comprehensively represent geometrical and thematic attributes, compositional and structural properties, spatial/topological relations between urban land cover patches (objects). Based on the evaluation of the importance of 26 spatial, thematic and topological variables in RANG, the random forest classification method is utilized to computationally infer and classify the urban land use in New Orleans into 7 types at the urban block level: single-family residential, two-family residential, multi-family residential, commercial, CBD, institutional, parks and open space, with an overall accuracy of 91.7%.

Date

1-15-2020

Committee Chair

Wang, Lei

Available for download on Tuesday, December 08, 2020

Share

COinS