Abstract. A realistic representation of snowfall in general circulation models (GCMs) of global climate is important to accurately simulate snow cover, surface albedo, high-latitude precipitation and thus the surface radiation budget. Hence, in this study, we evaluate snowfall in a range of climate models run at two different resolutions by comparing to the latest estimates of snowfall from the CloudSat Cloud Profiling Radar over the northern latitudes. We also evaluate whether the finer-resolution versions of the GCMs simulate the accumulated snowfall better than their coarse-resolution counterparts. As the Arctic Oscillation (AO) is the prominent mode of natural variability in the polar latitudes, the snowfall variability associated with the different phases of the AO is examined in both models and in our observational reference. We report that the statistical distributions of snowfall differ considerably between the models and CloudSat observations. While CloudSat shows an exponential distribution of snowfall, the models show a Gaussian distribution that is heavily positively skewed. As a result, the 10th and 50th percentiles, representing the light and median snowfall, are overestimated by up to factors of 3 and 1.5, respectively, in the models investigated here. The overestimations are strongest during the winter months compared to autumn and spring. The extreme snowfall represented by the 90th percentiles, on the other hand, is positively skewed, underestimating the snowfall estimates by up to a factor of 2 in the models in winter compared to the CloudSat estimates. Though some regional improvements can be seen with increased spatial resolution within a particular model, it is not easy to identify a specific pattern that holds across all models. The characteristic snowfall variability associated with the positive phase of AO over Greenland Sea and central Eurasian Arctic is well captured by the models.
Clouds damp the radiative impacts of polar sea ice loss