Study On Mixed Forecast Model Of Summer Precipitation In Jilin Province

This paper is an article that emphasize the exploratory and procedural,based on the Multivariate statistical analysis,the Lasso variable selection methods and Artificial neural networks and the other statistical research thoughts and modeling approaches,combined with many subjects such as geography,synoptic,atmospheric science and physics' theoretical knowledge,be aimed at exploring and establishing the short-term climate forecast model of summer precipitation in Jilin province.There were a large amount of experiments and analysis under constantly attempts and completions that have been developed in this article,finally,we devise a new modeling scheme that is utilizing Lasso variable selection techniques screens the main factors from the atmospheric circulations that influence the summer precipitation as mean as the main periodic sequences from the integrating extension series of the mean generating function which produced by precipitation sequence itself,then establishing the nonlinear mixed forecast model of double factors by the artificial neural network methods.The sum of rainfall amount June,July and August in northwest,south central and east-parts of Jilin province,respectively,as predicative object is carried out predicative experiment.The results show that the mixed forecast model of neural network based on double factors has more better prediction ability and physical foundation than the two kinds of single factor' regression prediction patterns of atmospheric circulation,ENSO meteorological index and mean generating function,and also than the double linear forecast model of mixed factors.