RADARSAT-2 Polarimetric SAR Data for Urban Land Cover Classification: A Multitemporal Dual-Orbit Approach

This research investigates multitemporal dual-orbit RADARSAT-2 polarimetric SAR data for urban land cover classification using an object-based support vector machine (SVM). Sixdate RADARSAT-2 high-resolution SAR data in both ascending and descending orbits were acquired in the rural-urban fringe of the Greater Toronto Area during the summer of 2008. The major landuse/land-cover classes include high-density residential area, low-density residential area, industrial and commercial area, construction site, park, golf course, forest, pasture, water and two types of agricultural crops. The results show that multitemporal SAR data improve urban land cover classification and the best classification result is achieved using data from all six-dates. However, similar accuracies could be achieved using only three-date data from both ascending and descending orbits with relatively longer temporal span. Combinations of SAR data with relatively short temporal span are observed to yield lower classification accuracy. Similarly, combinations of SAR data from either ascending or descending orbit alone yield lower accuracy than the combinations of ascending and descending data. The results indicate that the combination of both the ascending and descending spaceborne SAR data with appropriate temporal span are suitable for urban land cover mapping.