A mesh sensitivity study of lattice Boltzmann simulations for the flow over a three-dimensional hill

Airflow over complex terrain induces turbulent flows that must be accurately predicted for assessing wind farms. This study evaluates the capability of an in-house GPU-accelerated cumulant lattice Boltzmann method (LBM) code for simulating airflow over complex terrain. Computational accuracy and grid dependency were investigated through qualitative and quantitative comparisons of six mesh resolutions with experimental wind tunnel data on airflow past a three-dimensional isolated hill. Results indicate that an intermediate resolution, discretizing the hill height into 16.67 grids, provides the best compromise between accuracy and cost: the mean velocity profile converges within a computational time of 0.22 h, while satisfying wind-assessment metrics (Q-rate and FAC2) for both mean and turbulent quantities. Higher-order statistics and spectra continue to improve with finer resolutions, but from an engineering perspective, the intermediate resolution is the most practical choice, highlighting the high efficiency of the GPU-accelerated LBM framework for complex terrain wind simulations.