**2. Methods**

#### *2.1. Study Site and Data*

The Juvent wind farm in the Jura Mountains of Switzerland contains 16 wind turbines, twelve 2-MW Vestas V90's, and four 3.3-MW Vestas V112's, with a 95-m hub height. The turbines have 90-m and 112-m rotor diameters, respectively. The turbines are situated on two hills, Mont Soleil (alt. 1291 m) and Mont Crosin (alt. 1268 m), where surface cover varies from grassy to forested (Figure 1).

Wind measurements were taken at the nacelle of each turbine from the period 15 January– 11 February 2016. Data collected include wind speed, wind direction, power, yaw offset, and temperature. Each measurement was recorded as the mean over a 10-min interval. The standard deviation of wind speed over each interval was also recorded.

For the purposes of this evaluation study, only the predominant wind direction (i.e., 240◦ as shown in Figure 2) was simulated. Turbines 5–8 were excluded because they were shut down for replacement during the period of analysis. In addition, we focused on the cluster of nine turbines located on the hill of Mont Crosin (Figure 1). Thus, Turbines 9, 15, and 16 were not considered in the simulations because they are far away from the nine turbines and their influences on the flow in the area of interest is negligible. Data of the nine turbines were filtered to an average wind direction of 240 ± 3◦ and a wind speed range of 8–9 m/s as measured at Turbine 2, the farthest upstream turbine in the cluster. Wind speeds and turbine power outputs were normalized with the measurements at Turbine 2. The normalized results were then averaged over the filtered dataset because, in order to compare simulation results to observed data, we needed a single average measurement for wind speed and power at each turbine. Since Coriolis forces were assumed to be negligible in this study, normalization using linear scaling is valid [2]. Using normalized data from a certain range of wind conditions (hence, a larger dataset) allowed obtaining robust statistical results for a fair comparison.

**Figure 1.** Location of the study site in the Swiss Jura Mountains and numbering of the turbines on the Juvent wind farm (source: www.juvent.ch).

**Figure 2.** Wind rose at the Juvent wind farm in the Swiss Jura Mountains, 15 January–11 February 2016.

Wind fields in mountainous regions are highly turbulent and are strongly modulated by local, nonlinear interactions with multi-scale surface heterogeneities. The complex land features of interest include both mountainous terrain and heterogeneous vegetation. In this case study, the forest-grassland mosaics of the Jura mountains exhibit land cover whose effects on wind flow are difficult to model accurately. To apply the CFD tools under such complex surface conditions, we needed to feed them with high-resolution data of the relevant surface properties. The high-resolution data of the topography is directly used as input to the CFD tools to determine the surface elevation for the generation of the computational grid. The high-resolution data of the vegetation cover can be used to estimate

the surface roughness length in the similarity theory-based wall model and the parameters in the forest modeling. Forest can be modeled either directly through introducing additional forcing terms in the momentum equations or indirectly through the wall model with a high roughness length and displacement height.

In this study, elevation data at 25-m resolution were acquired from the Swiss topographical database through the www.geodata4edu.ch interface developed by the Swiss Federal Institute of Technology in Zurich (ETHZ). Land cover data at 25-m resolution were acquired from the CORINE land cover database, developed by the European Environment Agency, which classifies land cover into 44 different categories and provides the corresponding roughness length for each. Roughness length is a parameter of the vertical log-law profile that models the horizontal mean wind speed near the rough surface. It is equivalent to the height at which the wind speed theoretically becomes zero. As input to the model, we extracted from the elevation and roughness length maps a rectangular domain oriented towards the predominant wind direction (Figure 3). This ensures that the wind profile is allowed to develop over the same distance from every starting point along the inflow boundary. The dimensions of the domain were determined in a convergence test as 19 km × 5 km in the streamwise and spanwise directions, respectively, with 9-km spacing between the upstream border and Turbine 2. The elevation and roughness length presented in Figure 3 show that the Juvent wind farm is located in a highly-complex terrain. Patches with the roughness length value higher than 1 m are identified as forests, which are shown in dark red in the bottom panel of Figure 3.

**Figure 3.** Elevation (**top**) and roughness length (**bottom**) of the area of interest. The turbines are also presented in white circles.
