Detection of forest parameters using imaging spectroscopy

Kurzfassung

The main tree species in Bavaria is spruce, which is being strongly affected by the climate change, because of its anthropogenic influenced distribution in non-typical site ranges. Through climate change, biotic and abiotic factors, such as bark beetles, fungi, storm, snow and water stress are occurring more often and are stressing spruce in these unsuitable sites. Due to these changing conditions it is getting more important to detect forest types as a feature of its own, but also other tree species and spruce in particular. This thesis presents a spectral analysis of detecting deciduous and coniferous forest as forest type and independently European beech, European fir and Norway spruce as tree species. The analysis was carried out using hyperspectral HySpex VNIR and multispectral Worldview-2 images, each with 2 m ground resolution from a heterogeneous and stratified temperate forest in southern Germany. This type of forest exhibits characteristics of sustainable prospective temperate forests. A total of 2008 and 2009 reference spectra from HySpex VNIR and Worldview-2 were extracted and analysed with Principal Component Analysis for possible discrimination. A combination of eight uncorrelated bands, that were optimised for forest type discrimination and another eight bands, also optimised for tree species discrimination were extracted from HySpex VNIR imagery, using a Genetic Algorithm. The extracted eight bands from HySpex VNIR and the eight bands from W orldview-2 served as input for Linear Discriminant Analysis. The overall accuracies achieved from HySpex VNIR are 94.4 % for forest type and 88.3 % for tree species discrimination. Worldview-2 achieved an overall accuracy of 87.8 % for forest type and 86.7 % for tree species discrimination. The successful spectral based classification for forest type and tree species were transferred onto the corresponding images for spatial description of their occurrence. The promising results of this thesis confirms the advantage of airborne hyperspectral images for forest type and species detection. The transferability of this approach to spaceborne multispectral images with spectral information, relevant for vegetation purposes seems feasible.