Automated tree species mapping with high spatial resolution satellite images for large areas

Conventional methods in traditional forest management, such as stand estimation and forest inventories, are of only limited use to provide spatially detailed (i.e. at the sub-stand level) information on the true tree species distribution. This is particularly true if the stands are heterogeneous and rich in tree species diversity. Remote sensing is a very efficient technique for tree species mapping in such an environment, especially if high spatial resolution multispectral images are used. In this project, we use WorldView-2 data with 8 spectral bands and a pixel size of 0.5 m (panchromatic band) and 2 m (multispectral bands). In a pilot study (FFG Innovationsscheck), we could demonstrate that images from this sensor are suitable for detailed tree species mapping. An object-based approach using only the sunlit regions of the tree crowns was most successful. In the current project, the operational applicability of this approach is investigated by developing and evaluating different options for the technical implementation (feasibility study). The study area is located in the Austrian province of Burgenland. It is very rich in tree species diversity and therefore well-suited for this subject.