Hyperspectral remote sensing of vegetation species distribution in a saltmarsh

The availability of quality empirical data on vegetation species distribution is a major factor limiting the understanding, if not resolution, of many nature conservation issues. Accurate knowledge of the distribution of plant species can form a critical component for managing ecosystems and preserving biological diversity. Remote sensing is an important tool for mapping and monitoring vegetation. Advances in sensor technology continually improve the information content of imagery for airborne as well as space-borne systems. The unifying hypothesis of this dissertation was that vegetation associations can be differentiated using their hyperspectral reflectance in the visible to shortwave infrared spectral range. For this purpose the field reflectance spectra and airborne hyperspectral images of detailed saltmarsh vegetation types of the Dutch Waddenzee wetland were analyzed. Prior to analysis the field spectra were smoothed with an innovative wavelet approach which, compared to other techniques, showed the best trade-off between noise reduction and the preservation of spectral features. In the first stage of the analysis, the reflectance spectra of the vegetation types were tested for differences between type classes. It was found that, although vegetation spectra consist of similar detectable absorption features making them an important source of information about the biochemical constitution of vegetation, there are significant differences between vegetation types, both in absolute reflectance and in curvature. Using the airborne hyperspectral imagery, an alternative method was demonstrated that uses an expert system to combine airborne hyperspectral imagery with terrain data derived from radar altimetry. The accuracy and efficiency of production of the detailed vegetation map increased when generated by the expert system compared to those of a vegetation map produced by conventional aerial photograph interpretation. Lastly, the accuracy of classification for vegetation types was determined for the three data reduction techniques on both the field spectra and the imagery. Compared with the selection of individual bands, linear transformation of hyperspectral space into lower-dimensional improves the classification accuracy for vegetation classes, and therefore should be the preferred method. Therefore, the results confirm the main hypothesis that it is possible to differentiate vegetation using hyperspectral remote sensing.