Integrating Remote Sensing with Geospatial Analysis to Map and Interpret Vegetation Patterns in an Urban Environment

Vegetation is a fundamental component in urban ecosystems which provide human beings with various ecosystem services. Land changes due to rapid urbanization worldwide cause an unbalanced distribution of various vegetation types, which is difficult to map and characterize. Although many efforts have been made to examine the potential drivers of land changes, the specific processes leading to the spatial heterogeneity of vegetation patterns have not been fully understood. The overall objective of this research is to identify a remote sensing and GIS-based approach to help improve vegetation mapping in a complex urban area and to examine the factors shaping the observed vegetation patterns in the Atlanta metropolitan area as a case study site. The dissertation has gone through several major components. First, the algorithmic parameters affecting image classification accuracy by random forests are evaluated using both the classifier's accuracy and the map accuracy. Second, the performance of random forests for classifying complex landscape types relative to several popular classifiers including Gaussian maximum likelihood, support vector machines, and artificial neural networks is evaluated. Third, a multiple classifier system is further developed to derive a land cover map emphasizing various vegetation types for the study site. Last, the spatial patterns of various vegetation types are quantified using landscape metrics, and the factors shaping the observed vegetation patterns are explained from the biophysical, socioeconomic, and geographical perspectives at multiple observational levels using GIS and statistical analysis. Overall, this study has contributed to a better understanding of human and natural interactions in a complex urban environment through the combined use of remote sensing, GIS-based spatial analysis, and landscape metrics.