Mapping Urban Extent Using Satellite Radar Interferometry

Phase coherence between pairs of ERS SAR images is investigated as a method for mapping urban extent in South Wales, United Kingdom. Separability indices show that image pairs with time delays of greater than 2 months and baseline separations of less than 300 m can discriminate effectively between urban and non-urban land. Classification kappa coefficients greater than 90 percent are achieved, and there is evidence to suggest that a single coherence threshold is applicable for mapping urban extent in any similar landscape. Introduction Regular and up-to-date information on the extent of urban areas is primarily required for regional-scale planning purposes, such as mapping urban growth (Donnay, 1999; Weber, 2001; Bahr, 2001). Effective planning policy and appropriate resource management can only be accomplished through informed decisions, but even basic information on urban extent is often outdated, inaccurate, or simply does not exist (Barnsley et al., 2001). This is especially so within developing countries (Baudot, 2001). Current approaches to urban monitoring generally involve ground surveys and interpretation of aerial photographs, but these are not cost-effective solutions and are very difficult to implement. Thus, more efficient methods that are capable of automatically mapping urban areas are desirable. To this end, satellite remote sensing can be used to provide an objective and consistent view of urban areas, and SAR (synthetic aperture radar) in particular has the required coverage and revisit reliability for this application (Henderson and Xia, 1997). Urban areas are generally characterized by high SAR backscatter because of the predominance of singleand doublebounce scattering (Dong et al., 1997). However, it must be realized that the geometric relationship between azimuth angle and the orientation of the buildings strongly affects backscatter within towns and cities (Bryan, 1979; Hardaway et al., 1982). This may cause difficulties during the classification process because the same urban land covers can produce different backscatter intensities depending upon the azimuth angle of the sensor in relation to the built structures on the ground (Bryan, 1982). Recently, satellite radar interferometry has received a great deal of interest within the urban remote sensing community. The phase stability of anthropogenic structures between SAR images has led several authors to propose long time-scale phase correlation, or coherence, as a good measure of urban extent, and thus an appropriate tool for mapping urban change (Strozzi and Wegmuller, 1998; Usai and Klees, 1999; Strozzi et al., 2000). It has also allowed the exploitation of multiple phase measurements of point scatterers for the precise measurement of ground subsidence in urban areas (Ferretti et al., Mapping Urban Extent Using Satellite Radar Interferometry William Grey and Adrian Luckman 2001). Furthermore, digital elevation models derived by interferometry have been used to retrieve building height (Hepner et al., 1998; Gamba and Houshmand, 2000; Gamba et al., 2000). The degree of coherence between a pair of SAR images determines the quality of topographic or displacement information that can be retrieved by interferometry (Li and Goldstein, 1990; Rodriguez and Martin, 1992; Bamler and Hartl, 1998). It has also been shown to be of value in classifying properties of the land surface, for instance, in forest assessment as well as in urban analysis (Wegmuller and Werner, 1997; Askne et al., 1997). Coherence is influenced by a number of independent factors, including the time-delay between images, the difference in signals between images due to the different positions in space from which they were acquired, and other factors (Zebker and Villasenor, 1992). These other factors arise during data acquisition (e.g., thermal noise or differential atmospheric path delays) and processing (e.g., imperfect registration of the SAR images, causing mis-registration). Within urban areas coherence remains high even between image pairs separated by several years. Thus, when considering only the urban landscape, baseline decorrelation is the dominant factor, and coherence is only reduced slightly by temporal and other factors which are independent of baseline. In contrast, naturally vegetated surfaces are significantly influenced by temporal decorrelation and lose coherence within a few days or weeks as a result of growth, movement of scatterers, and changing moisture conditions. Hence, small-baseline, long time-scale coherence images (months to years) can be used to discriminate between urban and non-urban areas, allowing basic information on urban extent to be retrieved (Strozzi and Wegmuller, 1998). Moreover, a sequence of such observations can be used to automatically detect urban change. As well as being able to clearly delimit extensive built-up areas within coherence images, it is also possible to identify many hamlets, farms, and other isolated buildings within rural areas as very distinctive points of high coherence. Despite some studies illustrating the value of coherence in delimiting urban areas, there has been little attempt to quantify the accuracy of this technique. This paper provides a comprehensive analysis of the utility of ERS (European Remote Sensing) interferometric coherence data with a wide range of baselines and time-delays for distinguishing between urban and non-urban land covers over multiple cities within Wales and southwest England. Maps of urban extent derived from small-baseline, long time-scale coherence data and also backscatter data are validated. PHOTOGRAMMETR IC ENGINEER ING & REMOTE SENS ING September 2003 957 Department of Geography, University of Wales Swansea, Singleton Park, Swansea, SA2 8PP, United Kingdom ([email protected]). Photogrammetric Engineering & Remote Sensing Vol. 69, No. 9, September 2003, pp. 957–961. 0099-1112/03/6909–957$3.00/0 © 2003 American Society for Photogrammetry and Remote Sensing 03-911.qxd 8/7/03 5:21 PM Page 957