LAUSR

The strengths and weakness of the Twentieth-Century Reanalysis (20CR), the Pilot reanalysis of the twentieth-century (ERA-20C), a coupled twentieth-century climate reanalysis product (CERA-20C), and an ensemble of ten twentieth-century atmospheric model integrations (ERA-20CM), are examined for Antarctic snow accumulation based on 3265 Antarctic multi-year averaged surface mass balance observations and 79 ice core snow accumulation time series, which provides an independent estimate because they are not assimilated into the reanalyses and not used to force ERA-20CM. The ECMWF “Interim” reanalysis (ERA-Interim) and two regional climate models (RACMO2 and MAR) are also used as a complementary analysis. Despite the magnitude discrepancy between simulations and observations, large-scale spatial variability in multi-year averaged snow accumulation observations are reasonably well reproduced by all the twentieth century datasets. The four datasets capture a large fraction (> 40%) of the interannual variability in the ice core snow accumulation composite over the Antarctic Peninsula from 1901 to 2010. However, none of the twentieth century datasets alone is able to explain > 20% of variance in ice core records at the other Antarctic regions during the twentieth century. Even for the modern satellite era (from 1979 onwards), their performance for Antarctic snow accumulation is still poorer, relative to ERA-Interim, RACMO2 and MAR. Considerable inhomogeneities and spurious changes in atmospheric circulation are found in these datasets, and thus the precipitation minus evaporation/sublimation (P–E) changes and trends during the past 100 years are largely artificial over Antarctica.