Tropical cyclones similarity analysis based on manifold learning

Tropical cyclones (TCs) are one of the worst nature disasters. The analysis of tropical cyclones´ similarity is important in the TCs forecasting. In this paper, the similarities of TCs are studied in a fixed low dimension space. Both TC path and image are analyzed. Isometric Mapping (ISOMAP) is used to do dimensionality reduction. Several distribution characters, such as mean value, coefficient of skewness and kurtosis, are used as the features of clustering. The similarity of TCs in the low dimension space is homologous as in the high dimension space. Experiments also show that the similarity of TC paths is consistent with the similarity of TC images.