Land-cover Classification Using Radarsat and Landsat Imager for St. Louis, Missouri.

This paper presents the potential of integrating radar data features with optical data to improve automatic land-cover mapping. For our study area of St. Louis, Missouri, Landsat ETM+ and Radarsat images are orthorectified and co-registered to each other. A maximum likelihood classifier is utilized to determine different land-cover categories. Ground reference data from sites throughout the study area are collected for training and validation. The variations in classification accuracy due to a number of radar imaging processing techniques are studied. The relationship between the processing window and the land classification is also investigated. In addition, the Landsat images are fused with several combinations of processed radar features. The classification accuracies from the Landsat and radar feature combinations are studied. Our research finds that fusion of multi-sensor data improves the classification accuracy over a single Landsat sensor, although different processing techniques on radar images are required to obtain the best results. In our study, fusion of Landsat images and Radarsat feature combinations from a 13 × 13 entropy window, 9 × 9 data range widow, and 19 × 19 mean filter window achieves the highest overall accuracy improvement (10 percent) over the Landsat images alone. [ABSTRACT FROM AUTHOR]