Currently, a new generation of km-scale resolution global climate models are in development as the forthcoming phase of climate modelling. One such model is a 5-km version of the Icosahedral Nonhydrostatic Weather and Climate Model (ICON) developed jointly by Deutscher Wetterdienst (DWD) and the Max-Planck-Institute for Meteorology (MPI-M). Because of the high resolution, most parametrisations, such as that of convection and clouds, can be avoided.
Previous studies have identified substantial large-scale biases in model clouds over the Southern Ocean, affecting sea surface temperature and the Earth's albedo overall. Our aim is to quantify how well the high-resolution ICON model is simulating clouds in this region, particularly in light of the fact that subgrid-scale clouds are not parametrised in this model. This region is mostly dominated by boundary layer clouds generated by shallow convection, and these are problematic to observe by spaceborne lidar and radars, which are affected by attenuation by overlapping and thick clouds and ground clutter, respectively. Therefore, we choose to use a large set of ship-based observations conducted with ceilometers and lidars on board of RV Polarstern and other voyages. Altogether, we analyse about 1500 days of data from 31 voyages and 1 sub-antarctic station covering diverse longitudes of the Southern Ocean. To achieve a like-for-like comparison with the model, we use a ground-based lidar simulator called the Automatic Lidar and Ceilometer Framework (ALCF). We contrast the results with the ECMWF Reanalysis 5 (ERA5) and the Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2).
We show that the model underestimates the total cloud fraction by about 10%, with overestimation of cloud below 2 km, and underestimation of cloud above 2 km. The reanalyses also underestimate the total cloud fraction by about 20%. ERA5 overestimates cloud below 1 km but underestimates near-surface cloud or fog. In addition to lidar data, we compare radiosonde profiles acquired on the RV Polarstern voyages with ICON. Notably, the model exhibits smaller natural variability than observations, and its lifting condensation level tends to be higher. This might explain why cloud occurrence is peaking higher in the model (at 500 m) than in observations (at the surface).
The results imply that Southern Ocean cloud biases are still a significant issue in a km-scale resolution model, even though an improvement over the lower-resolution reanalyses is notable. More effort is needed to improve model cloud simulations in this fast-changing and understudied region. The advancement from convection and cloud parametrisation to cloud-resolving models might not solve this bias without an additional effort.
