Mapping urban areas using coarse resolution remotely sensed data

Identifying and anticipating the location, size and growth rate of urbanized areas is an important component to understanding, adapting to, and mitigating many aspects of global change. The main objective of this research is to improve understanding of the methodological, scale and validation requirements for mapping urban land cover from coarse resolution remotely sensed MODIS one kilometer data. Defining the extent of urban land is crucial, since knowledge of the size and spatial distribution of cities is important for regional and global environmental modeling as well as resource management and economic development planning. This research relies on the use of a supervised decision tree classifier, a nonparametric algorithm that has been shown to be effective for classifying noisy and incomplete data sets: a technique called boosting improves classification accuracy and provides a means to correct major sources of error using available prior information from the DMSP-OLS radiance calibrated nighttime lights data set. Results for North America indicate that the incorporation of DMSP-OLS data successfully improves urban classification results. Traditional as well as new measures of accuracy assessment demonstrate the effectiveness of the methodology for creating accurate maps of cities over large areas.