Population Density and Image Texture: A Comparison Study

The correlation between census population density and Ikonos image texture was explored. The spatial unit for the analysis was census blocks with homogenous land-use. Ikonos image texture was described using three methods: the graylevel co-occurrence matrix (GLCM), semi-variance, and spatial metrics. Linear regression was conducted to explore the correlation between image texture and population density. It was found that although correlation exists, its degree varies depending on the method used to describe image texture. The highest correlation is given by the spatial metrics method. This result suggests that the correlation between texture and population density is not strong enough to predict or forecast residential population. However, image texture does provide a base to refine census-reported population distribution using remote sensing. High-resolution satellite images therefore have the potential to support “smart interpolation” programs to estimate human population distribution in areas where detailed information is not available. Introduction Knowledge of the size and spatial distribution of human population in an urban area is essential for understanding social, economic, and environmental issues. Traditionally, census has been the primary source of information on population distribution and demographic characteristics. Due to the cost, frequency, and boundary designation problems associated with census (Openshaw, 1984), the utility of remote sensing for population estimation has been continuously explored since the 1950s. Various types of satellite imagery have been examined to study population distribution, including Landsat MSS (Iisaka and Hegedus, 1982), TM and ETM (Forster, 1985; Li and Weng, 2005), SPOT (Lo, 1995), and DMSP nighttime imagery (Dobson et al., 2000; Sutton et al., 2001). The successes of these studies vary, however one consensus reached is that for applications with small geographic extent, a spatial resolution of 0.5 to 5 m is necessary (Jensen and Cowen, 1999). Such a fine resolution was rarely available with previous satellite images. The advent of very high spatial resolution satellite images such as Ikonos renewed the interest in urban remote sensing including using remote sensing to estimate human population count and its spatial distribution. It has been suggested that an objective assessment of the current status Population Density and Image Texture: A Comparison Study XiaoHang Liu, Keith Clarke, and Martin Herold and future potential of urban remote sensing is necessary (Donnay et al., 2001). For urban population studies, Ikonos provides urban specificities comparable to those derived from low-altitude aerial photographs. The panchromatic imagery of 1 m spatial resolution enables the counting of individual dwelling units, while the multispectral imagery of 4 m spatial resolution clearly reveals the differences between residential neighborhoods. With the recognized limitations of medium-resolution satellite sensors (e.g., MSS, TM, and SPOT) and excitement about the new generation of high spatial resolution images like Ikonos, an examination of the utility of the new sensors is necessary. This paper intends to provide a discussion of the methodological challenges in conducting this line of research and present an examination of the correlation between Ikonos image texture and census population density. Remote sensing has long been used to estimate urban population and socio-economic parameters. Several surrogates derivable from remotely sensed imagery have been examined in the literature, such as the extent of an urbanized area, the spectral reflectance value, and the proportion of each land-use class. In this study, Ikonos image texture will be examined as a correlate of population density. The fundamental assumption is that neighborhoods with similar housing characteristics tend to have similar population density. Housing characteristics can be described by the size of the houses, the greenness, and other associated conditions. The interaction among these factors forms texture. Different housing conditions reveal different textures in remotely sensed imagery; consequently, if a relationship can be established between image texture and population density, texture can be use to inform studies of population count and its spatial distribution. Research in this paper will help to answer the following questions: Is there correlation between population density and Ikonos image texture?; Is the correlation strong enough to estimate population count?; Which method of texture description is most correlated with population density?. Answers to these questions will not only help to evaluate the utility of very high-resolution images for socioeconomic applications but also help to identify the challenges and future research needs in urban remote sensing. Remote Sensing and Studies of Human Population Research using remote sensing to study human population started in the 1950s and advanced in two interrelated PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING Feb r ua r y 2006 187 XiaoHang Liu is at San Francisco State University, Department of Geography & Human Studies, 1600 Holloway Avenue, HHS 279, San Francisco, CA 94132 ([email protected]). Keith Clarke and Martin Herold are at the University of California, Santa Barbara, CA 93106 ([email protected], [email protected]). Photogrammetric Engineering & Remote Sensing Vol. 72, No. 2, February 2006, pp. 187–196. 0099-1112/06/7202–0187/$3.00/0 © 2006 American Society for Photogrammetry and Remote Sensing 04-134.qxd 1/18/06 5:17 AM Page 187