Study Of Short Term Climatic Prediction Method Based On EMD And Ensemble Prediction

Climate system is a typical non-stationary system. However, climate observation data processing is usually completed under the assumption that time series are stationary, such as multi-step prediction of temperature or precipitation. Therefore they have lower accuracy. In order to improve this defect, this paper uses three time series prediction methods, each feature is studied and then the paper researches how integrating it with other two methods to improve current prediction models, the experimental data comes from anomaly percentage of February temperature data and summer precipitation data from 88 meteorological stations in Guangxi Zhuang Autonomous during 1957—2005, the conclusions and research details are as follows:(1) In data loading side, the new method named Empirical Mode Decomposition (EMD) is added. Combining it with normal prediction methods to presents a tentative improving measure. Through results comparing, the new algorithm can really reduce the instability of time series, and depressing its interference to the result.(2) In prediction result output side, adding Ensemble Prediction Technique, integrating several algorithms to presents ulterior improving measure, the experiments result indicate that relative to the directly prediction methods and normal prediction methods, the final prediction model which only use historical data can effectively predict the climate fluctuate trend and abrupt climate change in the case of multi-step prediction, and have better prediction accuracy.(3) Using the final prediction model to develop the prediction system, in response to the features of climate prediction and processing of climate time series, designing several system modules and their operating ways, have some certain generalization ability.