Investigations on methods of land cover classification of TM image in mountain area

The main purpose of this study was to develop a methodology for classification of Landsat imagery for mountain area cover type mapping. Single-stage classification and multi-stage iterative classification were evaluated to determine which classification of satellite imagery could be employed to obtain accurate land cover information in Mentougou district located in western Beijing with diverse topography. TM data used in the study consists of one quarter-scene acquired on 19 June, 2001. The use of ancillary information in the process of deriving thematic maps from satellite imagery was analyzed. Field surveyed data were not only used for the production of the vegetation map of Mentougou but for the ground information of the land covers in the study area Other ancillary data layers such as topography was used in the analysis. Five classification methods were used for classifying the TM data in the study: 1) single-stage classification, 2) single-stage classification with DEM analysis, 3) single-stage classification with PCA analysis, 4) iterative classification with band selection, and unsupervised classification. The accuracy of each classification is expressed as an error matrix from which the Kappa statistic and its large sample variance are derived. The results indicate that the iterative multi-stage classification approach was significantly better than the single-stage classification approach. And this organizational methodology for classification is feasible and reliable in mountain areas image classification. ? 2009 SPIE.EI