MONITORING OF FOREST CHANGES USING UNSUPERVISED METHODS - A CASE-STUDY FROM BOREAL FOREST ON MINERAL SOILS

An unsupervised clustering method was presented for monitoring rapid f orest changes such as cuttings, in large areas. The work was based on the hypothesis that most of the important changes in the forest canopy can be detected using space-born remote sensing information. Multitem poral Landsat TM data covering boreal forest were utilized. The result s showed that clustering of changes was not very accurate without prio r radiometric calibration. In this study a relative regression calibra tion was combined with Studentization of the spectral difference Featu res, The change detection accuracy at pixel level was not acceptable, therefore forest stands were used as classification units. It is propo sed that if forest stands are not available, spectral segments can be used as observation units for classification. The stand-level change d etection accuracy using the Kernel density linkage clustering method v aried from 87.6% with a three-year interval to 93.1% with a one-year i nterval.