Area-wide information on vegetation height is extremely valuable for many forestry applications. It enables the large-scale analysis and evaluation of various forest functions. For this reason, the Swiss National Forest Inventory (NFI) has been creating and publishing vegetation height models (VHM) for the whole of Switzerland since 2012. In this process, the vegetation height is modelled at a high spatial resolution of 0.5 m using photogrammetric methods from aerial stereo imagery. However, a countrywide update of the VHM is only possible every 6 years due to the acquisition strategy of the aerial imagery. Certain applications where the time factor plays an important role are limited by this update period. For example, a higher temporal resolution would be desirable for analyses on disturbances or changes at extreme locations.
Based on the optical satellite images of the Copernicus satellite mission Sentinel-2 and a Convolutional Neural Network (CNN), it is possible to generate an annual VHM for the whole of Switzerland. Since the Sentinel-2 images have a spatial resolution of 10-20 m, the vegetation height is modelled accordingly at a resolution of 10 m. In collaboration with the EcoVision group of ETH, the aim of the present project is to generate annual VHMs for Switzerland within the framework of the NFI. In the process, their accuracy is also to be assessed to be able to estimate the applicability of these VHMs for different use cases.