Efficient Water Area Classification Using Radarsat-1 SAR Imagery in a High Relief Mountainous Environment

It is important to determine quickly the extent of flooding during extreme cases. Even though SAR imagery with its own energy sources is highly applicable to flood monitoring owing to its sensitivity to the water area, topographic effects caused by local terrain relief must be carefully considered before the actual classification process. Since backscattering coefficients of the shadow area in high relief regions are very similar to those of the water area, it is essential to regard these areas before and after the classification procedure, although the process is a difficult and time-consuming task. In this study, efficient and economical methods for water area classification during floods in mountainous area are described. We tested five different cases using various synthetic aperture radar (SAR) image processing techniques, texture measures, and terrain shape information such as elevation and slope. The case whereby the SAR image was classified with the local slope information exhibited the best result for water area classification, even in small streams of different elevation categories. Consequently in mountainous areas, the combination of a SAR image and local slope information was the most appropriate method in estimating flooded areas. Introduction Synthetic Aperture Radar (SAR), an active sensor, transmits pulses of microwave and detects echo, which carries information about the surface. Due to relatively long wavelengths in microwave, radar signals are capable of penetrating clouds in the atmosphere and are independent of sunlight. These characteristics of SAR are particularly useful in monitoring floods over large areas, while accurate flood mapping using other methods is difficult since hydrologic instrument data and optical imagery are limited. The accurate delineation of flood extent provides important information that can help guide management decisions and provide necessary data for flood mapping applications. Although SAR data have been widely applied to these kinds of situations, some problems need to be solved before they can be put to actual use. Numerous investigations have been carried out to examine the capabilities of SAR sensors for wetland mapping and monitoring flooded areas (Imhoff et al., 1987; Hess et al., 1995; Pope et al., 1997; Brakenridge et al., 1998; Efficient Water Area Classification Using Radarsat-1 SAR Imagery in a High Relief Mountainous Environment Yeong-Sun Song, Hong-Gyoo Sohn, and Choung-Hwan Park Miranda and Fonseca, 1998; Alsdoforf et al., 2001; Townsend, 2001; Wickel and Jackson, 2001; Horritt et al., 2002). In previous works, since most of the areas studied involved flat terrain that did not cause serious radiometric distortions, it was relatively easy to delineate water extent. The works of Giacomelli et al. (1995), Birkett (1999), Liu et al. (2002), and Costa (2004) used SAR images to extract water areas through histogram analysis by setting the pre-defined threshold value. The threshold value, however, practically varies in every case, and it is not an easy to set the accurate threshold value to distinguish between water areas and non-water areas. In mountainous areas, the pre-defined threshold method may mistakenly classify non-water area into water area due to the more serious topographic effects caused by high terrain relief. The radiometric distortions depend strongly on the terrain and increase significantly in mountainous areas, in which the distortions should be corrected by a backscatter model for better classification results. Several studies have examined the influence of terrain relief on SAR images and proposed various correction procedures. van Zyl et al. (1993) and Ulander (1996) considered the inclination of the backscattering surface in azimuth and in range, in order to achieve more accurate radiometric corrections of topographic effects on SAR images. van Zyl et al. (1993) used the local incidence angle of the surface as a projection factor, while Ulander (1996) used the smallest angle between the surface normal and the image. Teillet et al. (1985) made use of an empirical cosine-based backscatter model, while Rauste (1989) examined the effect of topography on the imaging geometry and set up an empirical model of the backscatter variations. While an accurate Digital Elevation Model (DEM) is available, it is possible to correct topographic effects. However, this requires an enormous amount of time and complicated procedures. In very rugged terrain areas such as the Korean peninsula, new errors could manifest while removing topographic effects due to the uncertainties in the accuracy of elevation (Goering et al., 1995; Goyal et al., 1998; Sun et al., 2001; Bernier et al., 2002). Especially for radar shadows, since there is no signal to be normalized and no improvement can be expected, SAR images have limited use for flood monitoring in high terrain relief regions. PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING March 2007 285 School of Civil and Environmental Engineering, Yonsei University, 134 Shinchon-Dong Seodaemun-Gu, Seoul, 120749, Korea ([email protected]). Photogrammetric Engineering & Remote Sensing Vol. 73, No. 3, March 2007, pp. 285–296. 0099-1112/07/7303–0285/$3.00/0 © 2007 American Society for Photogrammetry and Remote Sensing CARRS-5 05/02/2006 3:56 PM Page 285