Global storm resolving models (GSRMs) represent the next generation of global climate models.One of them is a 5‐km Icosahedral Nonhydrostatic Weather and Climate Model (ICON). Its high resolutionmeans that parameterizations of convection and clouds, including subgrid‐scale clouds, are omitted, relying onexplicit simulation but necessarily utilizing microphysics and turbulence parameterizations. Standard‐resolution (10–100 km) models, which use convection and cloud parameterizations, have substantial cloudbiases over the Southern Ocean (SO), adversely affecting radiation and sea surface temperature. The SO isdominated by low clouds, which cannot be observed accurately from space due to overlapping clouds,attenuation, and ground clutter. We evaluated SO clouds in ICON and the ERA5 and MERRA‐2 reanalyzesusing approximately 2400 days of lidar observations and 2300 radiosonde profiles from 31 voyages and aMacquarie Island station during 2010–2021, compared to the model and reanalyzes using a ground‐based lidarsimulator. We found that ICON and the reanalyzes underestimate the total cloud fraction by about 10% and 20%,respectively. ICON and ERA5 overestimate the cloud occurrence peak at about 500 m, associated withunderestimated lower tropospheric stability and overestimated lifting condensation level. The reanalyzesstrongly underestimate fog and very low‐level clouds, and MERRA‐2 underestimates cloud occurrence atalmost all heights. Outgoing shortwave radiation is overestimated in MERRA‐2, implying a “too few, toobright” cloud problem. SO cloud and fog biases are a substantial issue in the analyzed model and reanalyzes andresult in shortwave and longwave radiation biases.
Publication Date: 2026-06-24