Classification of natural areas in northern Finland using optical remote sensing images and data fusion

The aim of this study was to make CORINE 2000 classification better in Northern Finland by making the classes of natural areas more separable using optical remote sensing images and decision based data fusion methods. Phenological time-series derived from MODIS images and ETM-image mosaic were classified using Maximum Likelihood classifier. Classification results were merged using different data fusion methods and their result compared. The used methods were not particularly successful, because the overall accuracy of CORINE 2000- classification, 52%, was better than the overall accuracies of spectral classifications and data fusion methods. The best data fusion methods were maximum joint a posteriori probability- classification and classification where a priori probabilities have been acquired from lower resolution classification.