Abstract: Bias correction for the 2 m air temperature from the T213 ensemble forecast product performed not good on dramatically cooling days using the original self-adaption Kalman Filter-typed decaying average bias correction method.In this study,the bias correction scheme w(i,p) is improved by redefining the decaying average weight w,with i representing station information and p representing synoptic process information,and the similarity w(i,p) method and the statistical w(i,p) method are further developed through optimizing effective extraction of historical information.The new improved bias correction methods have been evaluated.The result showed that the improved w(i,p) decaying average bias correction method has a better performance than the original method.The averaged root-mean-square (RMS) error of the 24-h forecast decreases by 0.15 ℃ for each member on dramatically cooling days.The statistical w(i,p) method has the best performance,with the averaged ensemble mean bias decreases by 2.54 ℃ compared with the w(i,p) decaying average bias correction method.