Research Of Developing Versus Nondeveloping Tropical Cloud Clusters Classification And Key Factors Mining Based On Decision Tree Algorithm

In order to improve the accuracy of tropical cyclogenesis forecasting, it is vital important to research on how tropical cloud clusters(TCCs) develop into tropical cyclones(TCs) and its climatic characteristics. TC genesis is insufficiently understood since most of previous studies focus on the time after its genesis. Although process models can be built to predict TC genesis, they generally fall short in capturing the intricacies of the underlying mechanisms. But data mining can reveal the rules hidden in the data, using the ability of its data analysis in combination with the kinetic theory of TCC development and the formation of tropical cyclones, prediction model is established to provide a new ideas of TC genesis research.In this paper, the development of TCC and TC genesis prediction rules are established by decision tree algorithm. The key factor which influence tropical cloud clusters activity were analyzed, obtained the following conclusions:(1) The TCCs develop into TCs was abstracted as a binary classification problem. Using the Cart and C4.5 algorithm respectively for 24h TC genesis events prediction, and compared with the discriminant analysis. Results indicated that C4.5 than CART algorithm with high accuracy, and training set and test set accuracy is 85.69% and 85.03% respectively. Through the analysis of the 11 kinds of TCCs environment flow found that decision tree algorithm can be used to identify the environmental flow characteristics of developing and nondeveloping TCCs.(2) 925hPa divergence and 700hPa relative vorticity selected by C4.5 algorithm is the most important factor which TCCs develop into TCs, and calculating the Box Different Index(BDI) found that 850hPa relative humidity is also very important. Through an analysis of the environment factor of developing and nondeveloping found that low-level relative vorticity, high-level and low-level divergence have significant differences, indicating that the TCC development of tropical cyclones are dependent on dynamics factor.In the rules of the C4.5 algorithm, the difference location of the TCC has great distinct and contrast found TCCs genesis productivity (DP) is highest in northwestern sub-region, lowest in southwestern sub-region of western North Pacific.(3)Sea surface temperature(SST) is an important factor affecting TCCs activity.During summer and autumn, the SST of developing and nondeveloping is not significant, which indicates that atmospheric environmental flow are the main factors influencing TCCs activity. So the abnormal changes of large-scale environmental flow caused by SST anomalies have important influence on the development of TCC. The correlation analysis is used between GP, Nino3 and El Nino Modoki index(EMI), the results indicated that GP is significantly positively correlated with Nino3 index in the southeastern sub-region of the WNP during summer.During fall,the GP is significantly positively correlated with Nino3 index in the southwestern and southeastern sub-regions. The EMI has a markedly positive(negative) correlation with the GP in the southeastern (South China Sea) sub-region during summer(fall).