Application Of Partial Least Squares Regression In Mult-modal Integrated Forecasting Of Water Vapor And Surface Air Temperature

The use of a new multi model integration method of partial least squares regression(PLS),its use can completely eliminate the multicollinearity features to improve the humidity and temperature of multi model integrated forecasting results.Partial least-squares regression in essence is a kind of super ensemble forecast,its advantage is completely eliminate the multicollinearity between each factor,can make each set of data are independent of each other does not interfere with each other,and all the other methods including eliminating deviation ensemble average(BREM),moving average is impossible to avoid multicollinearity,the multicollinearity between the various factors is easy to make the model is not stable,caused the error.PLS component in both retained the original argument more variance,and retained with the correlation of the dependent variable is bigger,to eliminate the original independent variables complex collinearity,regression models were set up still can fully reflect the relationship between independent variables and dependent variable.Based on the European Center for medium range weather forecasts(ECMWF),Chinese Meteorological Bureau(CMA),the Japan Meteorological Agency(JMA)and British Meteorological Bureau(UKMO)four center ensemble forecast result,build in 2012 multi mode(25-60 0N,60-150 0E)24 hours-168 hours forecast time(interval 24 hours)multi model,humidity and temperature,were used to eliminate the deviation of ensemble average(BREM),a simple set of average(EMN),super ensemble(SUP)and partial least squares regression(PLS)method for four kinds of ground temperature multi model integration,the root mean square error(RMSE)and Anomaly correlation coefficient(cor)to determine the effect of more mode of integration and for the prediction of a short course of cold.The four prediction results show that the partial least squares regression(PLS)best multi model integrated method,not only the four single mode is superior and compared to the other three methods showed better prediction performance,has a certain value and application prospect.