Abstract: This study evaluates the simulation ability of winter and spring Arctic Oscillation (AO) variability carried out with coupled climate models participating in the experiment “historical” organized by CMIP5. The features of AO during 1950-2000 in twenty eight models were analyzed and compared with those in the NCEP reanalysis data and twenty two CMIP3 models. The results show that, the significant increasing trend and oscillation cycle of winter and spring AO index can be reproduced by most models although the significant feature that winter/spring AO is in significant negative phase in the first 30 years whereas positive phase in the last 20 years is not shown up in the simulation, The multi-model ensemble has improved the effect of the simulations by CMIP5 models. As CMIP5, the CMIP3 models can not catch the feature that winter/spring AO is in significant negative phase in the first 30 years and positive phase in the last 20 years, what’s more, the significant increasing trend of winter and spring AO index is not shown up in the simulation by CMIP3 models, and there is not any improvement in the result of multi-model ensemble. There is a poor simulation for the variation cycles of AO index in the CMIP3 models. The oscillations with a cycle of 2-3 years in the winter AO index is reproduced by CMIP3 models, but the spring AO index simulated by CMIP3 models does not show the oscillation cycle of 4-5 years. Although the simulations are not good enough to catch all the significant features of winter and spring AO indices by CMIP5 models, there are some improvements not only for the increase trend but also for the change cycle in the simulations by CMIP5 models relative to CMIP3 models.

Key words: Arctic Oscillation (AO), CMIP5, CMIP3, EOF