Abstract:In this study,the empirical orthogonal function(EOF) was performed at the anomaly field of the 600-station winter mean temperature in China during the period of 1979-2012.Then,using the observed antecedent Arctic sea ice concentration(SIC) and Eurasian snow cover(SNC) data,the key areas where SIC and SNC anomalies in autumn have significant effects on the principal variation of following temperature in China are calculated,and based on those areas,the SIC and SNC indices are built.Next,the standard linear regression models which can be used to predict the mean winter temperature at individual stations are established,using one or two cryospheric predictor indices.Through the statistical cross-validation,the mean of the anomaly correlation coefficient(ACC) and root mean square error skill score(RMSESS) between the observed and predicted temperatures are used to quantitatively evaluate the predictive skill of cryospheric factors for the winter mean temperature in China.The results show that the skill of hindcasts is greatly different among regions between the single September SIC predictor and November SNC predictor.The SIC index has more noticeable skill on central north China,while the November SNC index has more noticeable skill on northeastern China.While hindcasts using both September SIC and November SNC predictors are better than the single on area and score,almost all stations except the Tibetan Plateau area show significant skill.The grid points with superior skill are centered on north-central,northeastern and northern China,where the regional average ACC is 0.58,and the method outperforms a climatological hindcast is 18.7%.The results obtained in this study suggest that it is very important to incorporate cryosphere variability in seasonal prediction systems.