Classification of urban areas at continental scales using remotely sensed data

This paper describes research to obtain continental and global scale maps of urban land cover from remotely sensed imagery, specifically utilizing newly available one kilometer data from the MODIS sensor. Defining the extent of urban land is crucial, since knowledge of the size and spatial distribution of cities is important on a number of fronts, from resource management to economic development planning to regional and global climate modeling. The algorithm used for this work is a supervised decision tree classifier, and the technique of boosting is exploited to improve classification accuracy and to provide a means to correct major sources of error using available prior information. First results for North America indicate that the incorporation of. ancillary information successfully improves urban classification results, resolving confusion between the urban and barren classes that normally occurs when only MODIS data is used