This dissertation includes two main parts. The first part proposes to employ the continuous, pixel-based landscape gradient models in comparison to the discrete, patch-based mosaic models and evaluates model efficiency in two empirical contexts: urban landscape pattern mapping and land cover dynamics monitoring. The second part formalizes a novel statistical model called spatially filtered ridge regression (SFRR) that ensures accurate and stable statistical estimation despite the existence of multicollinearity and the inherent spatial effect.
Results highlight the strong potential of local indicators of spatial dependence in landscape pattern mapping across various geographical scales. This is based on evidence from a sequence of exploratory comparative analyses and a time series study of land cover dynamics over Phoenix, AZ. The newly proposed SFRR method is capable of producing reliable estimates when analyzing statistical relationships involving geographic data and highly correlated predictor variables. An empirical application of the SFRR over Phoenix suggests that urban cooling can be achieved not only by altering the land cover abundance, but also by optimizing the spatial arrangements of urban land cover features. Considering the limited water supply, rapid urban expansion, and the continuously warming climate, judicious design and planning of urban land cover features is of increasing importance for conserving resources and enhancing quality of life.
This study seeks to determine the role of land architecture—the composition and configuration of land cover—as well as cadastral/demographic/economic factors on land surface temperature (LST) and the surface urban heat island effect of Phoenix, Arizona. It employs 1 m National Agricultural Imagery Program data of land-cover with 120mLandsat-derived land surface temperature, decomposed to 30 m, a new measure of configuration, the normalized moment of inertia, and U.S. Census data to address the question for two randomly selected samples comprising 523 and 545 residential neighborhoods (census blocks) in the city. The results indicate that, contrary to most other studies, land configuration has a stronger influence on LST than land composition. In addition, both land configuration and architecture combined with cadastral, demographic, and economic variables, capture a significant amount of explained variance in LST. The results indicate that attention to land architecture in the development of or reshaping of neighborhoods may ameliorate the summer extremes in LST.
Meanwhile, the emerging cyberinfrastructure rapidly increases our capacity for handling such massive data with regard to data collection and management, data integration and interoperability, data transmission and visualization, high-performance computing, etc. Cyberinfrastructure (CI) consists of computing systems, data storage systems, advanced instruments and data repositories, visualization environments, and people, all linked together by software and high-performance networks to improve research productivity and enable breakthroughs that are not otherwise possible.
The Geospatial CI (GCI, or CyberGIS), as the synthesis of CI and GIScience has inherent advantages in enabling computationally intensive spatial analysis and modeling (SAM) and collaborative geospatial problem solving and decision making.
This dissertation is dedicated to addressing several critical issues and improving the performance of existing methodologies and systems in the field of CyberGIS. My dissertation will include three parts: The first part is focused on developing methodologies to help public researchers find appropriate open geo-spatial datasets from millions of records provided by thousands of organizations scattered around the world efficiently and effectively. Machine learning and semantic search methods will be utilized in this research. The second part develops an interoperable and replicable geoprocessing service by synthesizing the high-performance computing (HPC) environment, the core spatial statistic/analysis algorithms from the widely adopted open source python package – Python Spatial Analysis Library (PySAL), and rich datasets acquired from the first research. The third part is dedicated to studying optimization strategies for feature data transmission and visualization. This study is intended for solving the performance issue in large feature data transmission through the Internet and visualization on the client (browser) side.
Taken together, the three parts constitute an endeavor towards the methodological improvement and implementation practice of the data-driven, high-performance and intelligent CI to advance spatial sciences.