Abstract: The differences of minimum and maximum air temperature between weather stations located in county and township of Anhui province were analyzed from October of 2008 to September of 2009. The results show that both differences are significant and it changes with the seasons. According to the forecast product of forecaster and the methods of the revised Barnes and the first-order Kalman filter, it forecasts the minimum and maximum air temperature in township. It suggests that the prediction accuracy rates are high using this method. The accuracy rates of the minimum temperature in the first 3 days are close to that in county and both differences are only in 1%, while the differences increase with the increasing of valid forecast time. For the maximum temperature, the accuracy rates that the forecast error is within 2 ℃ are still within 3 % compared with the forecast product in county. This method is better than the MOS method based on output products of the WRF model. The effects of the first-order Kalman filter after and before the revise show that the Kalman filter has a positive effect for township temperature in the first four days, while it has no effect or a negative effect from the fifth to seventh day. The forecast accuracy rates of the maximum temperature from the fourth to seventh days could be significantly improved after the turning synoptic processes are discriminated.

Key words: Temperature in country, Temperature in township, Barnes method, Kalman filter, Refinization