Date of Award

August 2016

Degree Type


Degree Name

Doctor of Philosophy



First Advisor

Changshan Wu

Committee Members

Mark D. Schwartz, Woonsup Choi, Alison Donnelly, Zengwang Xu


With rapid urbanization, impervious surfaces, a major component of urbanized areas, have increased concurrently. As a key indicator of environmental quality and urbanization intensity, an accurate estimation of impervious surfaces becomes essential. Numerous automated estimation approaches have been developed during the past decades. Among them, spectral mixture analysis (SMA) has been recognized as a powerful and widely employed technique. While SMA has proven valuable in impervious surface estimation, effects of temporal and spectral variability have not been successfully addressed. In particular, impervious surface estimation is likely to be sensitive to seasonal changes, majorly due to the shadowing effects of vegetation canopy in summer and the confusion between impervious surfaces and soil in winter. Moreover, endmember variability and multi-collinearity have adversely impacted the accurate estimation of impervious surface distribution with coarse resolution remote sensing imagery. Therefore, the main goal of this research is to incorporate temporal and spatial information, as well as geostatistical approaches, into SMA for improving large-scale urban impervious surface estimation.

Specifically, three new approaches have been developed in this dissertation to improve the accuracy of large-scale impervious surface estimation. First, a phenology based temporal mixture analysis was developed to address seasonal sensitivity and spectral confusion issues with the multi-temporal MODIS NDVI data. Second, land use land cover information assisted temporal mixture analysis was proposed to handle the issue of endmember class variability through analyzing the spatial relationship between endmembers and surrounding environmental and socio-economic factors in support of the selection of an appropriate number and types of endmember classes. Third, a geostatistical temporal mixture analysis was developed to address endmember spectral variability by generating per-pixel spatial varied endmember spectra.

Analysis results suggest that, first, with the proposed phenology based temporal mixture analysis, a significant phenophase differences between impervious surfaces and soil can be extracted and employed in unmxing analysis, which can facilitate their discrimination and successfully address the issue of seasonal sensitivity and spectral confusion. Second, with the analyzed spatial distribution relationship between endmembers and environmental and socio-economic factors, endmember classes can be identified with clear physical meanings throughout the whole study area, which can effectively improve the unmixing analysis results. Third, the use of the spatially varying per-pixel endmember generated from the geostatistical approach can effectively consider the endmember spectra spatial variability, overcome the endmember within-class variability issue, and improve the accuracy of impervious surface estimates.

Major contributions of this research can be summarized as follows. First, instead of Landsat Thematic Mapper (TM) images, MODIS imageries with large geographic coverage and high temporal resolution have been successfully employed in this research, thus making timely and regional estimation of impervious surfaces possible. Second, this research proves that the incorporation of geographic knowledge (e.g. phonological knowledge, spatial interaction, and geostatistics) can effectively improve the spectral mixture analysis model, and therefore improve the estimation accuracy of urban impervious surfaces.

Included in

Geography Commons