Evaluation of Worldview-2 Imagery for Urban Land Cover Mapping Using the Interimage System

Mapping of urban land cover using remote sensing technology has been widely explored, especially with the recent availability of high resolution images and object-based processing techniques. This study uses the InterIMAGE system and WorldView-2 orbital sensor imagery, two technologies which are new and still little explored in urban studies, to classify land cover in five test-sites near to the western section of Rodoanel Mário Covas, a ring-road that surrounds the metropolitan area of São Paulo, Brazil. The work hypothesis is: the spectral resolution increase of WorldView-2 imagery, compared to previous sensor systems, can improve the identification of urban targets, and consequently, improve the land cover classification. To evaluate the effects of the increase on spectral resolution of WorldView-2 system images, we simulated an image based on data from the QuickBird-2 sensor. Moreover, the framework for image classification InterIMAGE, which has been developed by PUC-RJ in cooperation with INPE, shows great potential for classifying complex urban areas. The proposed methodology is efficient to map the land cover in complex urban areas and the final classification of WorldView-2 images achieved an overall accuracy of 83% and a Kappa Accuracy Index of 0.81. The typical classification conflicts were solved, with a good identification of fifteen land cover classes. The results showed that the new spectral bands of the WorldView-2 were essential for discriminating some urban objects such as Ceramic Tile and Bare Soil, usually, difficult to be identified with other sensors of high spatial resolution, such as QuickBird-2. * Corresponding author.