Enabling the use of unstructured meshes for the Large Eddy Simulation of stable atmospheric boundary layers

Abstract. Modelling wind flows over complex terrain under varying atmospheric stability conditions is essential for improving our understanding of atmospheric boundary layer physics and its impact on wind energy systems. However, such simulations remain challenging due to the limitations of structured grids in representing complex geometries and the inherent difficulty of modelling the stable boundary layer, characterized by small-scale turbulent structures. These challenges necessitate the use of high-fidelity simulations with unstructured meshes, which offer greater geometric flexibility. Nevertheless, unstructured grids are rarely used in atmospheric simulations. This study establishes a baseline framework for the use of unstructured meshes in atmospheric boundary layer simulations, with particular relevance to complex terrain. The proposed solver is validated against two well-established benchmarks under neutral and stable stratification. For the neutral case, the Andrén benchmark, a 1.28 × 1.28 × 1.5 km³ periodic domain where the flow is driven by a large-scale pressure gradient, is considered. Results from structured and unstructured grids are in good agreement, with minor differences observed near the surface. Unstructured grids exhibit slightly higher friction velocities due to wall-proximal grid quality, but remain within the expected variability of existing studies. The solver is then applied to the GABLS1 stable boundary layer case, a 400 × 400 × 400 m³ domain with surface cooling. Both grid types capture the evolution of the SBL, with unstructured grids yielding higher surface heat fluxes – up to 14 % – resulting in a thicker boundary layer and noticeable differences in mean profiles and fluxes. A mesh refinement study confirms that a horizontal resolution of ∆x = 6.25 m is sufficient for accurate SBL representation with both mesh types. Overall, the results demonstrate that unstructured meshes are a viable and robust tool for atmospheric boundary layer modelling, capable of matching the accuracy of structured grids while offering the flexibility required for complex terrain. The minor discrepancies observed remain within the variability expected from model formulation choices. This work thus provides a foundational reference for future high-fidelity atmospheric simulations using unstructured grids, particularly in terrain-resolving contexts.