This proposal is motivated by the increasing need of spatially explicit information on tree species composition by different realms, former focus of the research reduced to case studies, and thus the lack of research for countrywide applications. In this project we will fill this gap and move from tree species classification on case study level to analysis at the country level. The goal is to develop a robust method to classify nine tree species with a spatial resolution of 10m for the whole of Switzerland. This approach is based on innovative remote sensing techniques and freely available data sets. It fully exploits the potential of multi-scale and multi-temporal Sentinel-1/-2 data from the ESA Copernicus program using deep learning classification algorithms. It is specifically designed to cope with heterogeneous forests, thus flexible regarding the choice of classification algorithm and economical regarding training data and it allows for regular updates and future adaptions.