Analysis of urban change detection techniques in desert cities using remote sensing

Rapid environmental and climate change at local, regional and global scales have been a major concern for scientists in various disciplines, such as geography, economics, environment, planning, and others. Land cover and land use changes have attracted attention because of the potential effects on erosion, increased run-off, water-balance, heat islands and climatological changes. An accurate knowledge of land use and land cover characteristics is essential to study their effects on human life and the environment. Urban expansion is an important type of land use and land cover change, since it is associated with population growth and economy. Remote sensing data has proved to be significant for monitoring and detecting urban change, and for providing essential information for future development. Change detection is the process of identifying differences in a land cover by observing it at different times. Different change detection techniques are usually compared to present the best change detection results for a particular application. This paper demonstrates the use of remote sensing to study the urban expansion in Dubai and Las Vegas from 1984 to 2010 using Landsat images, and assesses two different change detection techniques, post-classification comparison and principal component analysis. Urban change can be difficult to assess in desert cities using remote sensing because of the lack of contrast between highly reflective urban surfaces and highly reflective desert landscapes. The results of an accuracy assessment process in this research shows that a post-classification comparison technique provides more accurate results for quantifying urban sprawl in Dubai and Las Vegas than principal component techniques. Post-classification comparison techniques delivered more accurate results that ranged from 87.88% to 94.74%, while principal component analysis provided results with accuracies that varied from 46.94% to 83.92%. The reason for these variations in accuracy was the utilization of the unsupervised classification process on PCA multi-temporal images versus the supervised classification process.