Tree to Landscape Mapping of Forest Species Distributions from Hyperspectral Remote Sensing Data
The hyperspectral data acquired over the Injune study area (CASI, Hymap and Hyperion) have been used primarily for developing techniques for local descriptions of forests and to support upscaling of species distributions to generate community maps at the landscape level. Using the CASI data, an approach to the delineation of tree crowns has been developed using eCognition software. Tree species discrimination has also been established by extracting (for each delineated crown) averaged maximum values of spectral reflectance associated with specific pixels (e.g., band ratios) and other statistics (average of half and full crown) and applying stepwise discriminant analysis. Using this approach, accuracies exceeding 75 % were obtained in the classification of most tree species, although these decreased with the diversity. To link maps of crowns by species with structural information (e.g., height derived from LiDAR) and to also facilitate scaling of species information from the tree to the landscape level using the range of optical and radar remote sensing data available, automated registration procedures have also been developed. This paper outlines the approach developed for such registration and for classifying stands with a diversity of species, structures and growth stages from multi-spectral and hyperspectral data acquired at different spatial and spectral resolutions. The benefits of the developed techniques for regional mapping of tree species diversity and forest communities are presented and comparisons with existing approaches are made.