• deterministic prediction skill;
  • probabilistic prediction skill;
  • potential prediction skill;
  • EOF;


In this study, the superiorities of the super-ensemble for seasonal climate predictions are investigated based on the 500 mb geopotential height (GPH500) hindcasts produced by four Canadian atmospheric seasonal climate prediction models. The investigations are carried out mainly in two aspects: (i) a comprehensive evaluation of predictions for each grid point over the global domain by deterministic, probabilistic and potential prediction skill measures; (ii) the empirical orthogonal function (EOF) and the Maximum Signal-to-Noise (MSN) EOF analyses in the Northern Hemisphere. It is found that improvements of the super-ensemble are mainly due to the increase of ensemble size in the mid-high latitudes and the offsets of model uncertainties in the tropical regions. Measures of temporal correlation coefficient (CORR), Brier skill score (BSS) and reliability (REL) are more affected by the ensemble size; whereas the relative root-mean-square error (RRMSE) and resolution (RES) are sensitive to the offsets of model uncertainties. The first EOF mode of ensemble mean is similar to the most predictable pattern derived by the MSN EOF method, but the latter has the temporal evolutions more associated with the oceanic boundary forcing. The super-ensemble shows advantages in both EOF and MSN EOF analyses.