The European Centre for Medium-Range Weather Forecasts Reanalysis (ERA-40) and the National Centers for Environmental Prediction / National Center for Atmospheric Research Reanalysis (NCEP1) data are compared with Antarctic and other middle to high latitude station observations for the complete years of overlap, 1958-2001. Overall, it appears that ERA-40 more closely follows the observations; however, a more detailed look at the pre-satellite era reveals many shortcomings in ERA-40, particularly in the austral winter.

By calculating statistics in five-year moving windows for JJA, it is shown that ERA-40 correlations with observed MSLP and surface (2 m) temperatures are low and even negative during the mid-1960s. A significant trend in skill in ERA-40 is observed in conjunction with the assimilation of satellite data during winter, eventually reaching a high level of skill after 1978 that is superior to NCEP1. NCEP1 shows consistency in its correlation with observations throughout time in this season; however, the biases in the NCEP1 MSLP fields decrease significantly with time. Similar problems are also found in the 500 hPa geopotential height fields above the direct influences of the mountainous topography. The height differences between ERA-40 and NCEP1 over the South Pacific are substantial before the modern satellite era throughout the depth of the troposphere. The ability for ERA-40 to be more strongly constrained by the satellite data compared to NCEP1, which is largely constrained by the station observational network, suggest the differing assimilation schemes between ERA-40 and NCEP1 lead to the large discrepancies seen here. Thus, both reanalyses must be used with caution over high southern latitudes during the non-summer months prior to the assimilation of satellite sounding data.