Extrapolation of wind speed for wind energy - how to (?) for high nacelle heights

The past years have shown a rapid increase in the installation of new wind farms, despite the usual regulatory slow downs and social acceptance issues. Furthermore, existing wind farms have undergone repowering processes with some being repowered to newer, more efficient, and higher wind turbines. This repowering process included sometimes even the deinstallation of a number of wind turbines within a wind farm to avoid unwanted side effects such as wake inferences.  
This, however, poses some issues for the extrapolation of wind speed to hub heights higher than approximately 130m agl and adjustments to the power law and log law approaches may be necessary.
In Austria, a new wind atlas based on a concise set of 10 m meteorological observations of wind speed and direction, for wind energy applications and climate scenario downscaling, is currently being generated. With the idea of providing wind speed at different nacelles heights and including the requests of the wind industry, wind speed analyses at heights of up to 300m agl need to be provided. This requires thorough testing and adjusting of the classical methods as well as looking into machine learning methods for extrapolation (trained on e.g. the NEWA data).
In this work, a comparison of methods, including adjusting classical methods as well as testing ML methods for Austria, is shown. The methods are evaluated against both the NEWA data sets, the DANRA reanalysis, AROME operational, and AROME RUC data, was well as the ERA5 data. Furthermore, testing against open tall tower data such as the Cabauw tower or the FINO platforms is carried out.