The Research Of Atmosphere Data Based On Multilevel Statistical Model

Multilevel model is the main method of analyzing hierarchical data. Firstly, this paper describes the history of the development of multilevel model, and gived the basic form of multilevel model, then inferred the estimation of parameters and hypothesis test; when considering the estimation of parameters, it breaks the ordinary method of the estimation of the models, which describes how to using Gibbs sample method to estimate the parameters of the multilevel models. Secondly, describes the applied region of multilevel models, by transformed the data which is the rainfall of 23 cities in Hubei province from the year of 1995 to 2007, because of the difference of the region character, so the rainfall is varied; therefore, we can hold the series of rainfall as the level one, and hold the region character as the level two; then established a multilevel model which is concerned the cross-sectional study, and give the parameter's estimation of the model, holding the parameter's estimation as the development coefficient of the forecast model, and forecasting the rainfall, at the same time can test the model's significance, it shows that the model has the strong practice.Different year's rainfall of the same city can as a problem of repeated measurement, so we can establish a longitudinal study model and integrated the group explained variables into the model, which is used to explained the difference of the rainfall, forecasting the rainfall by using the longitudinal study model and taking the error analysis. This paper had established two models to forecast the rainfall already, and hold the moving average model to compared that two models, the result shows that the three models have strong practice, and can reflect the development trend of the rainfall, so in order to forecast the rainfall preferable, we hold the three models on a assembly model, which demands the error sum squares to minimum. Lastly, holding the assembly model compared two other models, according to the criterion of the average relative error, the result shows that assembly model has improvement comparing every single model.