Evaluation and diagnostics of the CERA-20C climate reanalysis ensemble

TitleEvaluation and diagnostics of the CERA-20C climate reanalysis ensemble
Publication TypeMiscellaneous
Year of Publication2018
AuthorDahlgren, P
Secondary TitleTechnical Memorandum

CERA-20C is a coupled ocean-atmosphere climate reanalysis produced by the European Centre for Medium RangeWeather Forecasts (ECMWF) and it covers the years 1901 to 2010. The atmospheric data assimilation is produced with a 4D-Var technique and assimilates surface pressure observations and wind speed at 10 metres over sea. It runs as an Ensemble of Data Assimilations (EDA) with 10 ensemble members. This report investigates different aspects of the behaviour of the atmospheric ensemble throughout the century.

By looking at how the analysis ensemble spread evolves with maps and time series it can be seen that the spread gradually decreases during the century as more observations are being assimilated into the system. It also shows that the surface observations have a substantial impact on the atmospheric state up to around 300hPa. This is especially noticeable in the northern hemisphere in the later part of the century.

Ensemble reliability studies have been performed for 2010 using observations of surface pressure as a reference for the truth. These show that the ensemble spread can indeed predict the error in the analysis ensemble mean state. It captures the spatial and temporal variations of the error but it is under dispersive (spread too small) in the extra-tropics. The larger the error, the larger the mismatch between spread and error; there is a conditional bias between the spread and the error. Compared to the operational EDA at ECMWF, the spread and error in CERA-20C are much larger and the operational EDA (which has 25 members) is more reliable.

Lastly, the ensemble mean and spread were compared to the climatological spread and variability by using a metric called Relative Entropy (RE). RE takes into account how both the ensemble mean and spread compares to the climate distribution and is used to assess the information content in the CERA-20C reanalysis. Maps from the beginning and the end of the century show that the information content increase with time and that the analyses in southern hemisphere in the beginning of century are close to the climate distribution.