To improve the effect of ensemble prediction, a 13member ensemble forecast of Masta was generated by using T106L19 model and breeding growth method(BGM),and the research of the application of ensemble prediction products from the results was carried out. Firstly, the pattern fields of ensemble forecast were clustered through variance analysis according to discrete distance between members. Through this method, the pattern fields are grouped into three different clusters, especially the several possible tracks and the proportions of every cluster member are provided. The 96 h forecast results were used as the basis of clustering in the previous method of discrete distance clustering. In comparison with the previous method, the results of variance analysis are better. Secondly, the evolvement of variables is categorized by using the method of systemic classifying and several possible typhoon landing positions can be concluded from clustering averaged plots. Based on these special points, the plumes plot with systemic classifying of variables can be used to analyze the sea level pressure of the special points. Through discussing the evolvement characteristics of these points, the prediction results of the landing time and position of typhoon are more similar to observations than the cluster average. In addition, boxwhisker plot has been presented. Several special ensemble members have been identified through the analysis of data beyond the main distribution of data. The information provided by the special points from boxwhisker plot can reduce the missing rate of severe weather event. The research results indicate that the effects of prediction and the application of ensemble prediction products were improved by using the method of the clustering basis determined by using variance analysis, systemic classifying and boxwhisker plot.