Potential of Hyperspectral Remote Sensing for Vegetation Mapping of High Mountain Ecosystems

Remotely sensed data is widely used in ecological applications because of its great advantages. These advantages are that the measures are being objective, repeatable and serve continuous data of the observed area opposed to traditional field survey based data collection (Zagajewskij et al. 2005). The remote observation of different vegetation species is important and quantitative and qualitative information can be derived by interpreting satellite/airborne images (Cjemg et al 2006). With the capabilities of hyperspectral sensors theoretically the possibilities are extended to derive more information with higher level of accuracy. In vegetation mapping and monitoring unfortunately this needs more sophisticated algorithms and techniques as traditional data processing techniques developed for processing of multispectral data are often fail because of the significantly more complex nature of hyperspectral imagery. Throughout the HYPER-I-NET project we aim to further extend the application possibilities of hyperspectral datasets therefore significant effort is being made in order to develop and test methodologies useful for hyperspectral data applications. Within the project we specifically study the vegetation related aspects of hyperspectral data processing such as land cover mapping and vegetation monitoring. In this field significant amount of research can be found within the technical literature but none of them is aiming the achievement of a generally applicable methodology that is useful for generic mapping purposes. Most of the techniques presented are highly specific for the specific application and the vegetative species that are being observed and mapped. In this paper we present the results of a comprehensive test of different hyperspectral data processing chain with the aim to identify the possibilities of moving forward to a generic data processing chain for vegetation mapping by means of hyperspectral imagery. During the research we built and tested certain data processing chains wile varying the methodologies used at different stages of the data processing system. As a following step an in-depth analysis of results were carried out aiming to find optimal solution for the given problem. The outcomes of the study shows that hyperspectral datasets can be successfully used in a generic vegetation mapping procedure but more careful design and implementation of classification system is required. Some of the standard data processing chains are selected and are being identified to be more suitable for vegetation mapping than the others. 2. INTRODUCTION AND BACKGROUND