Mesoscale to microscale coupling for determining site conditions in complex terrain

Accurate estimation of local site conditions, i.e. wind power resources and the turbulence impacting wind turbines turbines, is an important part of wind farm planning. In complex terrain, non-linear microscale models that account for atmospheric stability effects are needed for accurate site assessment, but by themselves they are unable to account for the changing large-scale weather conditions. Therefore, coupling to a mesoscale model, which can provide these variations to the microscale models, is required. This thesis describes, implements and validates with observed tall mast measurements a novel coupling strategy using the Weather Research and Forecasting (WRF) mesoscale model and the EllipSys3D URANS microscale model for wind downscaling. The coupling strategy is based on forcing the microscale model with momentum and temperature source terms extracted from the time-evolving mesoscale model simulation. These terms are included as source terms instead of the usual lateral boundary conditions. Two cases are presented for sites in simple and complex terrain. First, results from simple terrain cases are presented. It is shown that a Single-Column Model (SCM) version of microscale model forced by tendencies from WRF results in long-term wind statistics of comparable statistical accuracy to results using the WRF model itself. Using different Planetary Boundary Layer and Surface Layer schemes in the WRF model simulations shows that the SCM results tend to follow the WRF model results, while maintaining a statistically similar response near the surface. Second, the coupling method was used at the complex double hill site Perdigão. It was shown that for a simulation using 80 m grid spacing and a first-order accurate discretisation scheme, the coupled approach results in large improvements in wind statistics compared to downscaling with the WRF model with an innermost domain of 333 m grid spacing. At four masts situated on top of the two ridge tops, the mean biases in wind speed for a 32-day period were less than 3% at the top most anemometer on each mast, compared to mean errors of 8–12% and 17–25% for WRF domains of 1 km and 333 m grid spacing, respectively. Despite the encouraging results, a clear dependence on grid spacing and numerical methods was seen in the coupled model results. Uncertainties remain about possible double-counting of turbulent fluxes, and its impact on the simulated wind speed, when RANS turbulence closures are used for transient atmospheric modeling at fine resolution.