Simulation and Projection of the Arctic Oscillation in Winter Based on CMIP5 Models

On the basis of the NCEP/NCAR reanalysis data and CMIP5 19 model results, this work examined the performance of 19 CMIP5 models in the simulation of temporal variability and spatial pattern of the Arctic Oscillation (AO), and estimated the future changes of AO under two typical Representative Concentration Pathway scenarios (i.e., RCP4.5 and RCP8.5). Results show that most of the models can capture the basic structure of the AO mode, while MPI-ESM-LR and HadGEM2-AO can better simulate the overall AO mode than other models. However, some models overestimate the anomalous distribution on the Pacific side. Regarding the time series and temporal variability, the first mode of principal component (PC1) of the CMIP5 models basically can reproduce the weakening trend since 1950-1970, but the growing trend after 1970 is not obvious in the simulations. Nevertheless, the zonal index (ZI) sequence can simulate the trend in the two stages. On the whole, most of the PC1 and ZI sequences during 1950-2005 show a positive trend. More than half of the CMIP5 models can well simulate the high frequency cycles with 2-3 a; however, the simulations of 20-aquasi-periodic oscillation are poor. Among all the models, only CanESM2, CNRM-CM5, and GFDL-ESM2G can better simulate the ZI cycle inversion. It is found that the ZI sequence shows a significant upward trend under the RCP4.5 and RCP8.5 scenarios, and the interdecadal variation is obvious with three different stages, i.e., 2006-2039 and 2070-2001 correspond to two rising phases, while 2040-2069 is the slow descending phase. A majority of the simulation results reveals a positive trend, and more than 10 out of the 19 CMIP5 models have passed the test under both concentration scenarios.