ADDRESSING GEOGRAPHICAL CHALLENGES IN THE BIG DATA ERA UTILIZING CLOUD COMPUTING

Abstract

Processing, mining and analyzing big data adds significant value towards solving previously unverified research questions or improving our ability to understand problems in geographical sciences. This dissertation contributes to developing a solution that supports researchers who may not otherwise have access to traditional high-performance computing resources so they benefit from the “big data” era, and implement big geographical research in ways that have not been previously possible. Using approaches from the fields of geographic information science, remote sensing and computer science, this dissertation addresses three major challenges in big geographical research: 1) how to exploit cloud computing to implement a universal scalable solution to classify multi-sourced remotely sensed imagery datasets with high efficiency; 2) how to overcome the missing data issue in land use land cover studies with a high-performance framework on the cloud through the use of available auxiliary datasets; and 3) the design considerations underlying a universal massive scale voxel geographical simulation model to implement complex geographical systems simulation using a three dimensional spatial perspective. This dissertation implements an in-memory distributed remotely sensed imagery classification framework on the cloud using both unsupervised and supervised classifiers, and classifies remotely sensed imagery datasets of the Suez Canal area, Egypt and Inner Mongolia, China under different cloud environments. This dissertation also implements and tests a cloud-based gap filling model with eleven auxiliary datasets in biophysical and social-economics in Inner Mongolia, China. This research also extends a voxel-based Cellular Automata model using graph theory and develops this model as a massive scale voxel geographical simulation framework to simulate dynamic processes, such as air pollution particles dispersal on cloud.