Modelling the mountain boundary layer: Does higher resolution improve model performance?

The horizontal grid spacing of numerical weather prediction models keeps decreasing towards the hectometric range, where topography, land-use, and other static parameters are well-resolved. Still, models have to be evaluated over complex terrain, because it cannot be assumed that higher horizontal resolution automatically yields better model performance. In this study, we perform limited-area simulations with the ICON model across horizontal grid spacings (1 km, 500 m, 250 m, 125 m) in the Inn Valley, Austria. Simulations are ran with two turbulence schemes - a 1D parameterization and a 3D Smagorinsky-type scheme. We evaluate the model across scales with observations of the valley boundary layer from the CROSSINN measurement campaign. This allows us to investigate whether increasing the horizontal resolution automatically improves the representation of the thermally-induced circulation, surface exchange, and other mountain boundary layer processes. Results suggest that the valley topography is already well-represented at the kilometric range, but the simuations in the hectometric range show a more detailed representation of the vertical valley atmosphere structure and the up-valley flow. Across resolutions, the model struggles with the correct representation of interactions between larger and smaller scales. The two turbulence schemes show a similar performance, but the 3D Smagorinsky scheme simulates a delayed evening transition of the up-valley flow. It is argued that the major difference between schemes actually emerges from the different surface transfer schemes, and the choice of boundary layer parameterization is secondary.