Land-cover Classification Using Radarsat and Landsat Imagery.pdfVIP

Land-cover Classification Using Radarsat and Landsat Imagery for St. Louis, Missouri Heng Huang, Justin Legarsky, and Maslina Othman Abstract observation. They have limitations in acquiring cloud-free This paper presents the potential of integrating radar data imagery on a regular basis and difficulties in performing features with optical data to improve automatic land-cover spectral classification for certain types of land features mapping. For our study area of St. Louis, Missouri, Landsat (Ulaby et al., 1982). ETM and Radarsat images are orthorectified and co-regis- Compared to optical sensors, active microwave sensors tered to each other. A maximum likelihood classifier is can provide their own illumination of a scene under observa- utilized to determine different land-cover categories. Ground tion. The longer wavelengths enable penetration of atmospheric reference data from sites throughout the study area are conditions such as rain, sleet, fog, haze, smoke, precipitation, collected for training and validation. The variations in and clouds (Haack et al., 2000). Hence, an advantage of active classification accuracy due to a number of radar imaging microwave sensors such as Synthetic-Aperture Radar (SAR) is processing techniques are studied. The relationship between their ability to obtain images under various weather condi- the processing window and the land classification is also tions during both day and night (Curlander and McDonough, investigated. In addition, the Landsat images are fused with 1991; Goetz et al., 2000). Due largely to the wavelength, radar several combinations of processed radar features. The