An assessment of various image transformation techniques on Landsat TM Data for forest classification: a case study in kobernausserwald area, Austria

Abstract

LANDSAT Thematic Mapper data, collected over kobernausserwald area in Austria, 13 May 1993, were digitally analysed using different techniques (linear transformations) sunch as Normalized Difference Vegetation Index, Vegetation Index Transfrmed Vegetation Index, Principal Componet Analysis, Tasselled Cap Transformations and untransformed bands 2,3 and 4. The above techniques were used for digital forest classification of the area with maximum likelihood classifier and then assessed to know their accuracies. The assessments were made through two statistical methods: confusion matrix and kappa statistics. Confusion matrix method gave good result in tasseled cap transformations and principal component analysis with overall accuracy more than 85%. It was found conclusively that untransformed bands of LANDSAT TM data was quite useful and successful for forest classification. However it can be concluded the tasseled cap is the more recommendable one because of its high overall accuracy (>90%).