Date of Completion
Land cover classification; Land cover post-classification; Markov chain random field; Urban growth; Vertical urban growth
Daniel L. Civco
Field of Study
Doctor of Philosophy
The recently proposed Markov chain random field (MCRF) approach has great potential to significantly improve land cover classification accuracy when used as a post-classification method by taking advantage of expert-interpreted data and pre-classified image data. This doctoral dissertation explores the effectiveness of the MCRF cosimulation (coMCRF) model in land cover post-classification and further improves it for land cover post-classification and urban growth detection. The intellectual merits of this research include the following aspects: First, by examining the coMCRF method in different conditions, this study provides land cover classification researchers with a solid reference regarding the performance of the coMCRF method for land cover post-classification. Second, this study provides a creative idea to reduce the smoothing effect in land cover post-classification by incorporating spectral similarity into the coMCRF method, which should be also applicable to other geostatistical models. Third, developing an integrated framework by integrating multisource data, spatial statistical models, and morphological operator reasoning for large area urban vertical and horizontal growth detection from medium resolution remotely sensed images enables us to detect and study the footprint of vertical and horizontal urbanization so that we can understand global urbanization from a new angle. Such a new technology can be transformative to urban growth study. The broader impacts of this research are concentrated on several points: The first point is that the coMCRF method and the integrated approach will be turned into open access user-friendly software with a graphical user interface (GUI) and an ArcGIS tool. Researchers and other users will be able to use them to produce high-quality land cover maps or improve the quality of existing land cover maps. The second point is that these research results will lead to a better insight of urban growth in terms of horizontal and vertical dimensions, as well as the spatial and temporal relationships between urban horizontal and vertical growth and changes in socioeconomic variables. The third point is that all products will be archived and shared on the Internet.
Zhang, Weixing, "A Markov Chain Random Field Cosimulation-Based Approach for Land Cover Post-classification and Urban Growth Detection" (2018). Doctoral Dissertations. 1762.