Abstract A set of gridded, satellite‐based, precipitation products has been used to assess the performance of 46 Coupled Model Intercomparison Project (CMIP6) atmospheric‐only simulations and 5 reanalyses over the Southern Ocean (SO) on daily timescales, in terms of total precipitation and variance, frequency and intensity of wet days, and seasonal changes. Besides the expected “too frequent, too light” precipitation biases in most models and reanalyses over the region, our study reveals other undocumented features such as notorious peak shifts and shape changes in the frequency distributions from 35°–50° to 50°–65° bands, which do not occur in satellite estimates. We also evaluated snowfall over mid to high latitudes and found that models have a substantial bias in frequency and intensity of snow days (>1 mm/day). The fraction of snow to total precipitation is substantially smaller for AMIP, but due to a rainfall overestimation rather than a deficit in snowfall. The intercomparison of the 7 precipitation data sets that employ remote‐sensing observations (including GPCPv3.2 and IMERGv7) is characterized by large uncertainties, which are additionally discussed. Using previous studies based on in situ data over the SO as well as CloudSat, we can also infer biases in most satellite data sets.
Blanco et al. (Tue,) studied this question.