Limitations for multispectral classification of temperate forests-scaling problems of groundtruth and satellite data

This paper contributes to the development of a semi-operational forest cover classification method for monitoring the temperate European forest belt, using SPOT-XS or Landsat-TM on an example in Switzerland. Special emphasis is put on spatial resolution and scaling problems. Preprocessing of continuous binary groundtruth data, e.g. forest boundaries is examined. Digital forest cover classifications of SPOT-XS and Landsat-TM data were performed using an automatic parallelepiped algorithm. All data were geometrically corrected considering terrain induced errors, and radiometrically corrected considering variations of illumination. Both data sets were classified with varying pixel size between 20 m (HRV), 25 m (TM) and 100 m (both). For large scale applications where a 100 m cell size is preferred the effects of upscaling on classification accuracy is discussed.