THE EXPERIMENTAL FORECAST FOR AIRPORT TOTAL CLOUD COVER USING THE LINEAR AND NONLINEAR MODULES

摘要

It is necessary to develop a forecasting method which can resolve nonlinear problems because there are limitations in statistic forecasting methods mostly based on linear correlation when dealing with complicated nonlinear problems. Airport total cloud cover categorical forecast modules were developed applying the nonlinear methods including the back propagation neural network (BPNN), support vector machines (SVM) etc. and the multiple linear regression method, respectively, with the T213 medium-range numerical forecast products between 2003 and 2005. The experimental forecast was applied with the datasets in 2006 using the linear and nonlinear forecast modules, respectively and the forecast accuracy of them was compared. The results show that the forecast accuracy of two nonlinear forecast modules applied BPNN and SVM is better than that of linear forecast module applied multiple linear regression when using the same factor screening method. While the grading of total cloud cover adds to more than four classes, the forecast accuracy of SVM forecast module is better than the BPNN module.