**1. Introduction**

In recent years, wind flow simulations have gained popularity for wind energy applications, including wind resource assessment, wind power prediction, and wind turbine micro-siting [1]. Compared to field measurements, simulations offer high-resolution three-dimensional wind fields without the need for costly meteorological equipment. Originally, linear models such as the one implemented in the Wind Atlas Analysis and Application Program (WAsP) were used because of their efficiency and their sufficient accuracy over terrain with gentle slopes [2]. However, increased computational capacity combined with a need for more accurate predictions of wind flow over complex terrain have made Computational Fluid Dynamics (CFD) models both practical and necessary. Most simulations solve the steady Reynolds-Averaged Navier–Stokes (RANS) equations, which are time independent and which provide the statistics for wind velocity at each grid point [3]. Other CFD simulation techniques that have higher accuracy, but higher computational cost are also being developed to analyze wind flow patterns and wind farm performance. These time-dependent turbulence-resolving methods include Large-Eddy Simulation (LES) and Direct Numerical Simulation (DNS). LES uses a low-pass spatial filter to average out turbulence at small length scales. In this method, the computationally-expensive calculation of small turbulent structures is replaced by sub-grid-scale modeling. One example of an LES method for wind farm modeling can be found in Porté-Agel et al. [4]. DNS involves solving the full nonlinear Navier–Stokes equations, but is too computationally expensive to be applied in large applications such as wind farms.

WindSim, a CFD package for wind resource assessment and park optimization, has been used and evaluated in both industrial settings and academia. Several groups compared the results of WindSim to WAsP over complex terrain and found better performance in the CFD model WindSim [2,5,6]. Other studies validated the WindSim results against measurements without comparing to linear models [3,7–9]. Castellani et al. [8,10] evaluated turbine wake modeling in wind farms with complex terrain, compared results with on-site measurements, and studied the wake effects together with the terrain effects on the performance of wind farms. Cattin et al. [7] validated the use of WindSim over areas with heterogeneous land cover, but found that implementing a map of roughness lengths did not fully reproduce the effects of forested areas. Dhunny et al. [3] validated the application of WindSim in an island situation using two roughness lengths, one for land and one for sea. Waewsak et al. [9] applied WindSim to a wind resource assessment study in Thailand and found good agreemen<sup>t</sup> between simulation results and met mast measurements. Finally, Teneler [11] evaluated the forest model in WindSim and found that modeling the forest as a porous medium improved simulation accuracy in heterogeneous forested regions.

The aim of the present study is to perform a more comprehensive evaluation of the WindSim software taking into account the combination of three complexities: topography, heterogeneous surface cover varying between grassy and forested, and turbine wakes. To accomplish this, we applied WindSim to a case study of a wind farm in the Jura Mountains of Switzerland, for which field data are available. We first performed convergence tests for the simulation domain size and grid resolution. We then investigated WindSim's sensitivity to the forest model, the turbulence model, and the wake model.
