Abstract: Models of predicting the extreme minimum air temperature in five stations, i. e. Haikou, Baisha, etc., have deen set up using the Multiple Progressive Regression Analysis with Periodical terms. With these models, the trend of rubber tree chill disaster has been analysed.This method has taken into account the effects of predictors on the predicted variables, as well as the periodical changes of predicted variables themselves and the correlation periods of the predictors. The models give good results in both data fitting and predicting.