EXTRACTING USEFUL INFORMATION FROM THE OBSERVATIONS FOR THE PREDICTION BASED ON EMD METHOD

At present, the main proach to the climate prediction in practice such as the short-term and long-term climate prediction is using the statistical method to predict the climate in month scale, season scale and annual scale, respectively. Now the numerical model is capable to forecast weather up to 7 days, but there a re numerous difficulties in realizing the short-climate prediction by numerical integration due to the nonlinear/non-stationary effects of the climate system. In fact, most of the present statistical climate prediction methods (mainly includes empirical, mathematical and physical statistics methods) are based on the hypothesis that the system is stationary. However, the observations, in particul ar for the climate data, are often nonlinear/non-stationary and multihierarchical, which makes the prediction very difficult. Aiming at this problem, a new p rediction model is introduced, in which, firstly, using the empirical mode decom position the observation sequence are stationarized and a variety of intrinsic mode functions (IMF) are obtained; secondly the IMFs are predicted by the mean ge nerating function model separately; finally with the optimal subset regression model the part of predictions are used as new samples to fit the original series directly or step by step and a system of prediction equations are set up. The climate sequences prediction research shows that the individual IMF, especially the eigen-IMF, has more stable predictability than that of its sources. The trend of development in climate prediction lies in researching the mechanism and hier archy of the climate system, constructing the corresponding climate prediction model. An attempt has been accomplished in this paper. It is believed that the mo del proposed can open up a new effective way for the climate prediction or evaluation.