GIS-driven classification of land use using IKONOS data and a core national spatial information database

Integration of spatial information embedded in GIS databases with remotely sensed data is one of the most challenging issues in modern geo-information science. The broad availability of up-to-date satellite imagery and the rapid development of image analysis techniques have shifted the classification of remotely sensed data into an increasingly automated procedure. Compared to traditional mapping, automatic land-use classification has the advantages of lower cost, area-wide coverage, and the possibility of frequent updating. One approach to automatic classification is the GIS-driven methodology that integrates multispectral properties of satellite imagery with thematic and metric geospatial information by applying the theory of evidence. The practical implementation of this theory allows for the combination of evidence from mutually exclusive data sources, such as satellite imagery, digital air-photographs, or in situ spectral data and ancillary data extracted from the very same spatial data bases that are to be updated. The objective of this study was to perform and analyze a GIS-driven classification of land use based on IKONOS satellite data and the Israeli National GIS core spatial information database. The image objects (polygons) were classified using the land-use classes that are inherent in the National GIS. The knowledge about these land-use classes was formalized by intensity and shape parameters, captured from IKONOS spectral bands and the GIS spatial data as was defined in the land-use layer of the Israeli National GIS. The classification polygons were assigned with the probability of occurrence with one of analyzed training class types. A final classification was carried out by the Dempster's rule of evidence combination. The classification results (overall accuracy 82.7, kappa = 0.71) provide an indication of the utility of formalized knowledge for classification of land use. The proposed method could be useful for quality assessment and automatic updating of existing spatial databases.