Evaluation of CMIP5 and CMIP6 global climate models in the Arctic and Antarctic regions, atmosphere and surface ocean

Large efforts are engaged to model climate-ice sheet interactions in order to estimate Antarctic and Greenland ice sheets’ contribution to sea level in the next decades to centuries. Here we present a first-order evaluation of CMIP5 and CMIP6 climate models over both polar regions. We focus on large-scale atmospheric fields and surface ocean variables only. Our goal is to provide a first overview of climate model biases in polar regions, in order to use their outputs on an informed basis. We particularly target the use of climate model outputs for forcing ice sheet models and regional atmospheric models.

We consider 9 (non-independent) variables : 850 hPa and 700 hPa annual and summer temperature, annual integrated water vapor, annual sea level pressure, annual 500hPa geopotential height, summer sea surface temperature, and winter sea ice concentration; over the Arctic (> 50°N) and the Antarctic (<40°S) regions. We use the ERA5 reanalysis as a reference, but we also include 5 other reanalyses in the intercomparison in order to estimate uncertainty coming from this choice. We define two sets of metrics. The first set of metrics, called “scaled rmse”, is the spatial root mean square error (RMSE) of time-mean variables for each region, that we divide by the median RMSE among all CMIP models. The second set of metrics, called “implausible fraction”, is the portion of the region where the difference between time-mean CMIP model and time-mean ERA5 is greater than three times the local interannual standard deviation. We find a strong relationship between the two sets of metrics. In addition, using the implausible fraction, we find that CMIP variables are significantly more implausible in the Antarctic than in the Arctic. It might be because of badly resolved processes or because of higher decadal variability in the South. Further work should include estimates of decadal variability in the implausibility computation.