Spectral Mixture Analysis of the Urban Landscape in Indianapolis with Landsat ETM+ Imagery

This paper examines characteristics of urban land-use and land-cover (LULC) classes using spectral mixture analysis (SMA), and develops a conceptual model for characterizing urban LULC patterns. A Landsat Enhanced Thematic Mapper Plus (ETM+) image of Indianapolis City was used in this research and a minimum noise fraction (MNF) transform was employed to convert the ETM+ image into principal components. Five image endmembers (shade, green vegetation, impervious surface, dry soil, and dark soil) were selected, and an unconstrained least-squares solution was used to un-mix the MNF components into fraction images. Different combinations of three or four endmembers were evaluated. The best fraction images were chosen to classify LULC classes based on a hybrid procedure that combined maximum-likelihood and decision-tree algorithms. The results indicate that the SMAbased approach significantly improved classification accuracy as compared to the maximum-likelihood classifier. The fraction images were found to be effective for characterizing the urban landscape patterns. Introduction Urban landscapes are typically composed of features that are smaller than the spatial resolution of the sensors, a complex combination of buildings, roads, grass, trees, soil, water, and so on. Strahler, et al. (1986) described Hand L-resolution scene models based on the relationships between the size of the scene elements and the resolution cell of the sensor. The scene elements in the H-resolution model are larger than the resolution cell and can, therefore, be directly detected. In contrast, the elements in the L-resolution model are smaller than the resolution cells, and are not detectable. When the objects in the scene become increasingly smaller relative to the resolution cell size, they may be no longer regarded as objects individually. Hence, the reflectance measured by the sensor can be treated as a sum of interactions among various classes of scene elements as weighted by their relative proportions (Strahler, et al., 1986). Landsat Thematic Mapper (TM) or Enhanced Thematic Mapper Plus (ETM+) images with a nominal 30 meter spatial resolution are attributed to L-resolution model. These data are often considered too coarse for mapping the components of urban environments. As the spatial resolution interacts with the fabric of urban landscapes, a speSpectral Mixture Analysis of the Urban Landscape in Indianapolis with Landsat ETM+ Imagery Dengsheng Lu and Qihao Weng cial problem of mixed pixels is created, where several landuse and land-cover (LULC) types are contained in one pixel. Such a mixture becomes especially prevalent in residential areas where buildings, trees, lawns, concrete, and asphalt can all occur within a pixel. Mixed pixels have been recognized as a problem affecting the effective use of remotely sensed data in LULC classification and change detection (Fisher, 1997; Cracknell, 1998). Fisher (1997) summarized four causes of the mixed pixel problem, i.e., (1) boundaries between two or more mapping units, (2) the intergrade between central concepts of mappable phenomena, (3) linear sub-pixel objects, and (4) small sub-pixel objects. When mixed pixels occur, pure spectral responses of specific features are confused with the pure responses of other features, leading to the problem of composite signatures (Campbell, 2002). The low accuracy of LULC classification in urban areas is largely attributed to the mixed pixel problem. For example, the traditional per-pixel classifiers, such as maximum-likelihood classifier (MLC), cannot effectively handle complex urban landscapes and the mixed pixel problem. When unsupervised classification is applied to densely populated suburban metropolitan areas, the mixed pixel problem becomes exaggerated. Trees on lawns are confused with forest classes. Lawns are similar to pasture and recreation, and pavement is common in high-density residential and commercial/industrial areas (Epstein, et al., 2002). In practice, accurate classification results are a prerequisite for many environmental and socioeconomic applications, such as urban change detection (Chen, et al., 2000; Ward, et al., 2000), urban heat islands (Lo, et al., 1997; Quattrochi, et al., 2000; Weng, 2001), and estimation of biophysical, demographic, and socioeconomic variables (Lo, 1995; Thomson and Hardin, 2000). Improving LULC classification accuracy has been an important theme in remote sensing literature. Different approaches have been used to improve urban LULC classification or change detection accuracies. These approaches include incorporation of geographic data (Harris and Ventura, 1995), census data (Mesev, 1998), texture features (Myint, 2001; Shaban and Dikshit, 2001), and structure or contextual information (Gong and Howarth, 1990; Stuckens, et al., 2000) into remote sensing spectral data, use of expert systems (Stefanov, et al., 2001; Hung and Ridd, 2002) and fuzzy classification (Zhang and Foody, 2001), use of multisensor data such as merged radar and TM data (Haack, et al., 2002), merged SPOT and TM data (Gluch, 2002), and merged P H OTO G R A M M E T R I C E N G I N E E R I N G & R E M OT E S E N S I N G September 2004 1 0 5 3 Dengsheng Lu is with the Center for the Study of Institutions, Population, and Environmental Change, Indiana University, Bloomington, IN 47408 ([email protected]). Qihao Weng is with the Department of Geography, Geology, and Anthropology, Indiana State University, Terre Haute, IN 47809 ([email protected]). Photogrammetric Engineering & Remote Sensing Vol. 70, No. 9, September 2004, pp. 1053–1062. 0099-1112/04/7009–1053/$3.00/0 © 2004 American Society for Photogrammetry and Remote Sensing 03-019.qxd 8/10/04 17:01 Page 1053