The Discovery of Tropical Cyclone Dynamics in Western North Pacific through Data Mining.

摘要

The booming population and property values of coastal areas, together with changing climatic conditions, mean that tropical cyclones (TCs) are coming to wreak increasing levels of damages on human beings. Data mining methods (e.g., C4.5 algorithm) are used to tease out hidden information on the patterns, regularities or structures of cyclonic behaviors including TC recurvature, TC landfall, post-landfall TC analysis, and intensity related to landfall and recurvature from the archived TC data in western North Pacific, including South China Sea. A web-based system is built for TC analysis and prediction to combine the outputs of data mining and dynamic models. The primary contributions of this thesis may be summarized as follows:;1. A decision tree representing TC recurvature with 18 leaf nodes is built through the C4.5 algorithm based on the potential variables that may influence TC recurvature. Eighteen rules are derived from the tree by following a pathway from the root node to each leaf node. All these rules can be explained from the meteorological perspective. The chosen variables and cutting values in the decision tree (e.g., 123 °E and 130 °E) identify the particular importance in estimating and predicting of controlling the TC recurvature.;2. A decision tree stipulating TC landfall with 14 leaf nodes is built by C4.5 algorithm selecting potential variables and cutting values to build the tree. Therefore, 14 rules are derived from the tree by tracking the path from the root node to each leaf node. The rules comprised by chosen variables and cutting values are interpreted by TC movement theories. The classification result of the decision tree is verified according to 2010 TC best track data. The chosen variables and cutting values of the tree provide references for predicting TCs' landfall.;3. The analysis of the historical post-landfall tracks of TCs over China unravels three clusters using Finite Mixture Model (FMM)-based clustering. The clusters are interpreted by the concepts of geopotential height, water vapor supply and baroclinic energy from mid-latitude circulations. This form of cluster analysis for the first time identifies the clusters hidden in the TCs that made landfall over the Chinese coast.;4. By revisiting the intensity change associated with recurvature, the study determined that TCs are prone to peak their intensities near their recurvature. We also found that the recurving points of TCs peaking in intensity after recurvature tend to be located to the southeast of those of TCs peaking in intensity prior to recurvature. The recurving TCs sustain their maximum intensity for significantly longer than straight-moving TCs. The TCs tend to weaken significantly from landfall--24h to landfall+24h.;5. The study builds a novel web-based system for TC analysis and prediction. This system is web-based and has great potential for technology transfer. Its user-friendly interface offers a multi-criteria query system for tracking and visualizing TC movements on the basis of mined data and dynamic modeling. It can playa significant role in TC analysis and prediction for professionals and general users.;6. This thesis focuses on how large-scale circulation and meteorological variables surrounding TC centers influence TC motion. TCs actually interact with large-scale circulation. At the same time, the TC structures and internal convections also exert impacts on TC motions. With the development of advanced observing equipment (e.g., radar, satellites and other forms of reconnaissance), the variables related to internal TC dynamics can be collected to complement the analysis of large-scale circulation.