Seasonal Sensitivity Analysis of Impervious Surface Estimation with Satellite Imagery

Numerous approaches have been developed to quantify the distribution of impervious surfaces using remote sensing technologies. Most of these approaches have been applied to data from a single time period, typically in the summer season (June to September). Presently, it is not clear whether there is an optimal time for impervious surface estimation with these methods. In this paper, the seasonal sensitivity of impervious surface estimation is examined. In particular, Landsat TM/ETM imagery for four different seasons has been acquired for the environs of Franklin County, Ohio. Two impervious surface estimation methods, spectral mixture analysis and regression modeling, are used to test for seasonal variations. Results indicate that the summer image provides better accuracy with the spectral mixture analysis method, while consistent accuracies are obtained for all four seasons with regression modeling. Introduction Impervious surface refers to any material that water cannot infiltrate. Typical impervious surfaces include building rooftops, streets, highways, parking lots, and sidewalks, all of which are major components of urban infrastructure. Therefore, impervious surfaces have been found to reveal essential information about built-up areas, and can be utilized to quantify urban development and land-use intensity. Ridd (1995) proposed the vegetation-impervious surface-soil (V-I-S) model to parameterize the biophysical composition of urban environments, and illustrated the relationship between impervious surface distribution and urban land-use types. Applying the V-I-S model, Rashed et al. (2001) and Lu and Weng (2004) quantified the extent of urban and suburban development and indicated that urban land-use classification accuracy can be significantly improved with impervious surface information. Yang et al. (2003b), Rashed et al. (2005), and Xian and Crane (2005) analyzed urban growth rates and patterns, and suggested that impervious surface information serves as a better alternative than traditional measurements of urban growth. Besides quantifying urban extent and land-use, imperviousness has been utilized to assess adverse influences of urbanization on water quality, urban climate, air quality, and natural habitat (Dougherty et al., 2004; Schueler, 1994; Weng et al., 2004), and it has been listed as a key indicator Seasonal Sensitivity Analysis of Impervious Surface Estimation with Satellite Imagery Changshan Wu and Fei Yuan for the health of water and terrestrial ecosystems by U.S. Environmental Protection Agency (USEPA, 2003). In addition to its role in evaluating the effects of urbanization on natural environments, impervious surface information also serves as an important factor in urban socio-economic studies. With the help of impervious surface information, Wu and Murray (2005) estimated detailed population distribution in Columbus, Ohio; Yu and Wu (2004) incorporated impervious surface fraction in understanding population segregation patterns in Milwaukee, Wisconsin. Also, Yu and Wu (2006) evaluated the influences of impervious surface distribution on housing prices in Milwaukee, Wisconsin. Because of the influencial role that impervious surface plays, the generation of impervious surface data has become an emerging area of interest to both scientists and decision makers. Traditionally, impervious surface areas have been quantified based on photographic or survey methods, in which impervious surfaces are digitized manually from high-resolution air photographs or survey maps. Although relatively accurate, these methods are labor intensive and time consuming. Consequently, various techniques have been developed in recent years for the automatic estimation of impervious surface fraction from medium-resolution remote sensing imagery such as Landsat Thematic Mapper (TM) and SPOT satellite data. In summary, major methods include spectral mixture analysis (SMA), regression modeling, regression trees, artificial neural networks (ANN), and subpixel classification. Spectral mixture analysis involves modeling a mixed spectrum as a combination of spectra for pure land-cover types, also called endmembers, such as vegetation, impervious surface, and soil. Applying the spectral mixture analysis method, Phinn et al. (2002) estimated impervious surface distribution with endmembers chosen from aerial photos. Wu and Murray (2003) implemented a constrained linear SMA to generate an impervious surface distribution for Columbus, Ohio, and found that the impervious surface fraction can be estimated by a linear model of low and high albedo endmembers. Further, Wu (2004) proposed a normalized spectral mixture analysis (NSMA) to achieve a better estimation accuracy of impervious surface distribution. Instead of utilizing a single set of endmembers, Rashed et al. (2003) developed a multiple endmember spectral mixture model to quantify impervious surfaces. Regression modeling involves estimating impervious surface distribution using the greenness component generated from a Tasseled Cap transformation applied to PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING Decembe r 2007 1393 Changshan Wu is with the Department of Geography, University of Wisconsin-Milwaukee, Bolton Hall 462, P.O. Box 413, Milwaukee, WI 53201 ([email protected]). Fei Yuan is with the Department of Geography, Minnesota State University-Mankato, 7F Armstrong Hall, Mankato, MN 56001. Photogrammetric Engineering & Remote Sensing Vol. 73, No. 12, December 2007, pp. 1393–1401. 0099-1112/07/7312–1393/$3.00/0 © 2007 American Society for Photogrammetry and Remote Sensing 05-133.qxd 2/11/07 10:37 Page 1393