The classification of tropical cyclones(TCs) is significant to obtaining their temporal and spatial variation characteristics in the context of dramatic-changing global climate. A new TCs clustering method by using K-...The classification of tropical cyclones(TCs) is significant to obtaining their temporal and spatial variation characteristics in the context of dramatic-changing global climate. A new TCs clustering method by using K-means clustering algorithm with nine physical indexes is proposed in the paper. Each TC is quantified into an 11-dimensional vector concerning trajectory attributes, time attributes and power attributes. Two recurving clusters(cluster A and E)and three straight-moving clusters(cluster B, C and D) are categorized from the TC best-track dataset of the western North Pacific(WNP) over the period of 1949-2013, and TCs' properties have been analyzed and compared in different aspects. The calculation results of coefficient variation(CV) and Nash-Sutcliffe efficiency(NSE) reveal a high level of intra-cluster cohesiveness and inter-cluster divergence, which means that the physical index system could serve as a feasible method of TCs classification. The clusters are then analyzed in terms of trajectory, lifespan, seasonality, trend,intensity and Power Dissipation Index(PDI). The five classified clusters show distinct features in TCs' temporal and spatial development discipline. Moreover, each cluster has its individual motion pattern, variation trend, influence region and impact degree.展开更多
基金National Key Research and Development Program of China(2016YFC0401903)National Natural Science Foundation of China(51722906,51679159,51509179)Tianjin Research Program of Application Foundation and Advanced Technology(15JCYBTC21800)
文摘The classification of tropical cyclones(TCs) is significant to obtaining their temporal and spatial variation characteristics in the context of dramatic-changing global climate. A new TCs clustering method by using K-means clustering algorithm with nine physical indexes is proposed in the paper. Each TC is quantified into an 11-dimensional vector concerning trajectory attributes, time attributes and power attributes. Two recurving clusters(cluster A and E)and three straight-moving clusters(cluster B, C and D) are categorized from the TC best-track dataset of the western North Pacific(WNP) over the period of 1949-2013, and TCs' properties have been analyzed and compared in different aspects. The calculation results of coefficient variation(CV) and Nash-Sutcliffe efficiency(NSE) reveal a high level of intra-cluster cohesiveness and inter-cluster divergence, which means that the physical index system could serve as a feasible method of TCs classification. The clusters are then analyzed in terms of trajectory, lifespan, seasonality, trend,intensity and Power Dissipation Index(PDI). The five classified clusters show distinct features in TCs' temporal and spatial development discipline. Moreover, each cluster has its individual motion pattern, variation trend, influence region and impact degree.
