According to the current state of knowledge, forests with a higher diversity of tree species offer more ecosystem services than forests with lower diversity of tree species. Knowledge about the distribution and abundance of tree species is necessary to better understand their contributions to ecosystem function and the provision of ecosystem services. Sweden is a large country, and therefore, it is important to map the distribution and abundance of different tree species over large areas. Traditionally, mapping of tree species has been done using optical remote sensing data from airborne systems or satellites, combined with simple algorithms like the Maximum Likelihood method. The results have been satisfactory, but these approaches are not suitable for larger areas.

State-of-the-art deep learning algorithm

This project utilises state-of-the-art deep learning algorithms in combination with radar and optical data from the Sentinel-1 and Sentinel-2 satellites to map tree species in southern Sweden. The microwave radiation emitted by the Sentinel-1 satellite can penetrate through the tree canopies and provide information about the structure of the trees. The sensor aboard the Sentinel-2 satellite can capture sunlight reflected from the Earth across a wide spectrum of wavelengths. This includes wavelengths from the "red edge" portion of the electromagnetic spectrum, which is sensitive to the chlorophyll content in vegetation and thus helps differentiate between different tree species. In this way, these two satellites complement each other and help improve tree species identification.

By utilising the latest deep learning algorithms in combination with these complementary data sources, there is an opportunity to further enhance the determination of tree species. Overall, this method leads to an operational process where satellite data and deep learning alone are sufficient to regularly provide maps of tree species distribution in Sweden.