Testing reanalysis data sets in Antarctica: Trends, persistence properties, and trend significance

The reanalysis data sets provide important sources for investigating the climate in Antarctica where stations are sparse. In this paper, we compare the 2 m near-surface temperature data from five major reanalysis data sets with observational Antarctic stations data over the last 36 years: (i) the National Centers for Environmental Prediction and the National Center for Atmospheric Research Reanalysis (NCEP1), (ii) NCEP-DOE Reanalysis 2 (NCEP2), (iii) the European Centre for Medium-Range Weather Forecasts Interim Reanalysis (ERA-Interim), (iv) the Japanese 55 year Reanalysis (JRA-55), and (v) the National Aeronautics and Space Administration Modern-Era Retrospective-analysis for Research and Applications. In our assessment, we compare (a) the annual and seasonal trends obtained by linear regression analysis, (b) the standard deviation around the annual trends, (c) the detrended lag-1-autocorrelation C(1), (d) the Hurst exponent.. that characterizes the long-term memory in a record, and (e) the significance levels of the warming/cooling trends. We find that all five reanalysis data sets are able to reproduce quite well the long-term memory in the instrumental data. In contrast, C(1), which is needed as input for the conventional significance analysis, shows fully erratic behavior. The observational warming/cooling trends in East and West Antarctica are not reproduced well by all reanalysis data sets, in particular, NCEP1, NCEP2, and JRA-55 show spurious warming trends in many parts of East Antarctica, even in those parts where cooling has been observed. In contrast, the standard deviation around the trends is quite well reproduced by all reanalysis data sets. In the peninsula where the station density is quite high, the performance of the reanalysis data is considerably better. It is remarkable that all reanalysis data sets as well as the observational data show (under the assumption of a long-term persistent process) that in the considered time period since 1979 the warming in the peninsula is not significant with p values well above 0.1.