Global Urban Area Mapping in High Resolution Using Aster Satellite Images

We present development of automated algorithm for mapping global urban area in high resolution using ASTER satellite images and coarse-resolution urban area maps. The algorithm consists of two steps: classifying pixels of ASTER satellite images into urban or non-urban by Learning with Global and Local Consistency (LLGC) technique; and integration with existing urban area maps using logistic regression. We implemented the algorithm and demonstrated it against 340 scenes of ASTER satellite images. LLGC trimmed up 500-m-resolution clusters of urban area into 15-m-resoluton clusters. However accuracy assessment on LLGC result showed 75% user’s accuracy, 41% producer’s accuracy, 94% overall accuracy and 0.50 kappa coefficient, indicating LLGC had considerable misclassifications due to similarity in surface reflectance among non-vegetative land cover. To complement the misclassifications, we integrated LLGC result with existing urban area maps. Accuracy assessment on result of the integration showed 74% user’s accuracy, 43% producer’s accuracy, 94% overall accuracy and 0.51 kappa coefficient, indicating that the results were more accurate than LLGC result and existing urban area maps. We concluded our method would improve global urban area map not only in terms of spatial resolution, but also in that of accuracy.