Accuracy of Classification of Remotely Sensed Data in Vegetation Mapping Process – A Case Study of Croatia

Vegetation is a fundamental variable that affects and links many parts of physical environments. Changes in vegetation cover have significant effects on basic processes (biogeochemical cycling, soil erosion) and thereby on biodiversity. In this study the thematic accuracy of classification of remotely sensed data in vegetation mapping process has been computed. The investigation area included the Nature Park l Žumberak – Samoborsko gorjer . Tree thematic maps with different minimum mapping unit (2.25 ha, 9 ha, 25 ha) were used. Accuracy assessment was carried out at two levels: Level I – 2 categories: nonforest and forest habitats, and Level II – 5 categories: anthropogenic habitats, nonforest vegetation, fir forests, coniferous forests, oak forests. The results show that overall accuracy at Level I ranges from 83% to 94%, and that KHAT values range from 63.54% to 87.06%. At Level II the results show that overall accuracy ranges from 49% to 71.50%, and that KHAT values range from 29.89% to 61.74%. This study shows that the accuracy of classification of remotely sensed data depends on both the minimum mapping unit and the number of categories used in classification scheme.