A Subgrid Topography–Surface Radiation Parameterisation in ICON for NWP

Incoming surface radiation in mountainous terrain is strongly modulated by topography. Direct shortwave radiation can be blocked by shadow casting from surrounding terrain and depends on local slope and aspect. Diffuse shortwave and longwave radiation are also affected by terrain, although typically to a lesser extent. These terrain-induced effects on surface radiation influence surface energy balance components such as ground temperature and snow cover, as well as near-surface atmospheric variables including 2 m air temperature and humidity. Through land–atmosphere interactions, such impacts can propagate vertically and affect boundary-layer processes, valley wind systems, and convection.

In atmospheric models such as ICON, topographic influences on surface radiation are not explicitly resolved. The radiative transfer scheme operates in the vertical direction only and assumes horizontally aligned grid cells, thereby neglecting the influence of local and surrounding terrain on radiation. To address these limitations, ICON includes a parameterisation that accounts for terrain effects on direct shortwave radiation on the model grid scale.

At MeteoSwiss, ICON is operated for numerical weather prediction at horizontal resolutions of 1–2 km. At these scales, fine-scale topographic features present in the high-resolution input dataset, available at 30 m, are substantially smoothed and thus not considered in the grid-scale topography-surface radiation parameterisation. To capture these unresolved effects, we implemented a subgrid-scale extension of the topography–radiation parameterisation in ICON. The required static input fields are derived using a high-performance ray-tracing algorithm. The new parameterisation is verified against station observations of global radiation, sunshine duration, and 2 m air temperature. Results show clear improvements, particularly in mountainous valleys and during winter months when solar elevation angles are low. In addition, a comparison of simulated snow cover with a satellite-derived product also reveals improved performance on seasonal timescales.