This study evaluates a numerical weather prediction model as a tool for wind resource assessment in complex terrain and how the simulations are affected by the selection of initial and boundary conditions and various grid resolutions. Two global reanalyses, ERA-Interim and ERA5, and four grid resolutions, 27 km, 9 km, 3 km and 1 km, have been considered. The simulations have been compared to hub-height wind measurements. The ERA5 forced simulations were found to provide improved wind speed results. The simulations show clear improvements in terms of lower error when the grid spacing is reduced from 27 km via 9 km to 3 km. From 3 km to 1 km, the error is not further reduced. However, the 1 km simulations tend to better reproduce the mean wind features and show defined areas with higher and lower wind speeds, providing useful information for wind resource mapping in complex terrain.