Subalpine and alpine vegetation classification based on hyperspectral APEX and simulated EnMAP images. Issue 7 (3rd April 2017)

• Record Type: Journal Article
• Authors: Marcinkowska-Ochtyra, Adriana; Zagajewski, Bogdan; Ochtyra, Adrian; Jarocińska, Anna; Wojtuń, Bronisław; Rogass, Christian; Mielke, Christian; Lavender, Samantha
• Is Part Of: International journal of remote sensing. Volume 38:Issue 7(2017)
• Page Start: 1839
• Page End: 1864
• Publication Date: 2017-04-03
• Subjects: Remote sensing -- Periodicals; Télédétection -- Périodiques; 621.3678
• DOI: 10.1080/01431161.2016.1274447
• Languages: English
• ISSNs: 0143-1161

Abstract

ABSTRACT: The characterization of vegetation is a very important ecological task, especially in sensitive mountain areas, as alpine regions often respond to small short-term variations of abiotic and biotic components as well as long-term global changes. Spatial techniques, such as imaging spectroscopy, allow for detailed classification of different syntaxonomic categories of vegetation and their status. Based on the Airborne Prism Experiment (APEX) and simulated Environmental Mapping and Analysis Program (EnMAP) data, this study focused on subalpine and alpine vegetation mapping in the eastern part of the Polish Karkonosze National Park (KPN). The spatial resolution of APEX (3.12 m) enabled the classification of 21 vegetation communities, which was generalized into eight vegetation types. These types were identified on scaled-up APEX data, as both 252 bands from most of the spectral range and a spectrally reduced dataset of 30 minimum noise fraction (MNF) transforms, and compared to the simulated (30 m spatial resolution) EnMAP data using test areas extracted from the field survey derived reference non-forest vegetation map. After preprocessing, a pixel purity index (PPI) was calculated using the MNF image and then the training and validation pixels were selected with Support Vector Machine classification of vegetation communities carried out using different kernel functions: linear, polynomial, radial basis function, and sigmoid. The classification accuracy was obtained for 21 base classes, and the best result was achieved by using the linear function and 252 bands (overall accuracy (OA) of 74.39%). The next step was to classify the eight generalized vegetation types, and the OA for the APEX data reached 90.72% while EnMAP reached 78.25%. The results show the potential use of APEX and EnMAP imagery in mapping subalpine and alpine vegetation on a community and vegetation-type scales, within a diverse ecosystem such as the Karkonosze National Park.