A multi-strategic approach for land use mapping of urban areas by integrating satellite and ancillary data

A methodology for land use mapping of urban areas by using satellite imagery and ancillary data (road network and census data) was developed. The integration of ancillary data was sequentially done to allow a cost/benefit analysis. In the analysis of the satellite imagery, special emphasis was put in the development of contextual operators for (1) discriminating pixels that have different land uses but have similar spectral characteristics, and (2) identifying land use classes that cannot be identified at a pixel level. A road network map was integrated with the satellite imagery for stratifying the study area into urban and rural areas. This stratification allowed the application of different algorithms to each stratum for an effective improvement of the land use mapping accuracy. In addition, population and habitation census data were used for a refinement of land use classes discrimination. The methodology was tested with a SPOT image to generate a map with the CLUSTERS nomenclature for the Area da Grande Lisboa, Portugal. In the final map, 22 land use classes were identified with an overall accuracy of 88% (Kappa index). Most of these 22 classes were identified with user´s and producer´s accuracy larger than 85%