LAUSR

Nighttime light (NTL) and OpenStreetMap (OSM) have been used increasingly to delineate boundaries between urban and nonurban areas. However, systematic comparisons of how well such data can be used for identifying urban land for cities with different development levels are extremely limited. In this study, NTL data from the Defense Meteorological Satellite Program/Operational Linescan System, road data and points-of-interest data from the OSM are carefully selected as main data sources, and further applied for urban land extractions from Chinese cities at different development levels. Approaches being adopted for extractions include the support vector, optimal threshold, sudden-jump, head/tail break, and densi-graph methods. Results show that the overall accuracy of urban land extracted from OSM data is significantly higher than that from NTL data. Averaged overall accuracies (AOAs) of urban land extractions from OSM data are 90%, while AOAs from NTL data are only 76%. Accuracies of urban land extractions experience a decline during increasing city development levels. Averaged balanced accuracies (ABAs) for high-developed cities are the lowest (about 65%), while ABAs for mid- and low-developed cities are comparable (71% and 72%, respectively). Also, significant differences between accuracies of urban land extractions by different methods are not observed in this case. Further, it is suggested that OSM is a robust data source for extracting urban land from cities at different development levels.