Assessment On The Prediction Of The Arctic Oscillation In CMIP5 Decadal Experiments

The Arctic Oscillation(AO)has an important impact on the Eurasian winter climate and extreme weather events,and is also a strong signal reflecting climate change.Coupling model is an objective tool for describing the climate system.How to improve the skill in predicting AO in decadal experiment is still a critical issue.Based on the observations,the results of started annual decadal experiments in 9 CMIP5 models and historical experiments in the BCC_CSM1.1,we assess prediction skill for AO,analyze the bias-correction statistical method and sources of the model biases.The main conclusions are as follows:(1)The spatial correlation coefficient of AO mode simulated by the decadal experiments is higher than that in the historical experiments.Compared with historical prediction,the correlation coefficient of the monthly and winter AO index is higher in the decadal experiments.In particular,the correlation coefficient of the monthly AO index between the decadal simulations and the observation exceeds 0.1 significant level.Furthermore,the periodicity of the monthly AO index are achieved only in the decadal experiments.Hence,the hindcasts skill for AO is robust when the initial state is initialized by sea surface temperature data.(2)In the 1-9 year and 4-year average forecasts,the winter AO mode and variability which are simulated by the decadal experiments one year ahead with 9 CMIP5 models,are closest to the observed results.Notably,the winter AO characteristics are reasonably predicted at one year lead time with GFDL-CM2.1,EC-Earth and GEOS-5 models.(3)The skillful prediction of the positive anomaly center of AO one year ahead is closely related to the skill in reproducing the autumn sea temperature(sea ice)and its association with winter AO.Due to the underestimation of the coupling between the stratosphere and the troposphere,the intensity of the simulated Arctic center is weaker than that of observation.In addition,the biases of winter AO centers simulated by decadal hindcasts one year ahead,which are consistent with the deflection of AO-related storm track activity anomalies,are also affected by the skill in predicting the climatological mean of the storm track activity.The robust skill for the positive feedback of synoptic eddy on AO plays an important role in reasonably reproducing the winter AO pattern with decadal experiments.The results suggest that the development of simulated sea temperature(sea ice),stratospheric circulation and storm track activity can further improve the hindcasts skill for AO in the future decadal experiment.(4)There are discrepancies between the AO reflected by the decadal hindcasts and the observation.The winter AO bias-correction methods for removing North Pacific dominant mode(NPM)and multiple linear regression have their own advantages.The biases in the North Pacific center of AO pattern can be reduced with the former method.And the latter method can comprehensively improve the prediction of three centers of AO.