An improved locally adaptive supervised classification method for the land cover mapping using Earth observation data has been developed. Detailed a priori probability databases has been implemented to describe the spatial extent of land cover types. This ensured a possibility to involve a number of auxiliary thematic products into classification of cloud-free MODIS composite images. These products have been derived using specialized methods including satellite observation time series analysis (e.g. burnt area data), GIS analysis (e.g. tundra and steppe areas) and using expert knowledge concerning global spatial biome distribution and structure.

The computing system developed to perform the locally adaptive supervised classification algorithm has been optimized for multiply repeated classifications procedure. This option is important to ensure high computational efficiency when there is need for product improvement as result of introduction of new thematic or regional expertise, or to eliminate misclassification by reference data modification and satellite images reclassification. The data processing performance was increased due to selection of areas of interest among the entire territory and therefore reduction of computation time. It made possible to update the classification product versions regularly. Classification control and data access facilities have been implemented in the web interface.

The first version of the Vegetation Cover of Russia map has been completed. It is based on Terra-MODIS data for the year 2005 and has spatial resolution of 250 meters. Monthly of cloud and snow free image mosaics have been used for classification. As the sources of reference data the SPOT-Vegetation based GLC2000 map, the USSR forest map, Circumpolar Arctic Vegetation Map and others have been involved. The legend of the new map consists of 21 land cover classes, including 8 types of different forest types.

Fig. The vegetation cover map of Russia