Improving Prediction Skills of Winter Temperature in China using Summer-Autumn Surface and Stratosphere Signals

Seasonal prediction of regional climate remains a challenge. This study seeks to improve regional climate predictability in China using comprehensive antecedent observations. The predictive skill for winter mean temperature in China is estimated by evaluating statistical cross-validated hindcasts made using simple linear regression models with four summer-autumn variables, including October-November Eurasian snow cover, July-August-September North Pacific sea surface temperature (SST), September Arctic sea ice and September 100-hPa geopotential height, which significantly influence the following Northern Hemisphere winter atmospheric circulation. Maximum covariance analyses are first performed between each anomaly field of these variables and the 600-station winter mean temperature in China during the period of 1979-2012, and indices are then built based on key areas where these variables have significant impacts on winter temperature in China. Standard linear regression models to predict mean winter temperature at individual stations are built using one or all four predictor indices. Hindcasts using a single predictor show noticeable skill over most areas, but using all predictors almost all stations except in the Tibetan Plateau area show significant skill. The anomaly correlation coefficient ranges from 0.55 to 0.70, and this method outperforms a climatological hindcast by up to 30-50% over most stations of eastern China. Results here suggest that it is important to incorporate cryosphere variability in seasonal prediction systems.