Evolution Map: Modeling State Transition of Typhoon Image Sequences by Spatio-Temporal Clustering
The purpose of this paper is to analyze the evolution of typhoon cloud patterns in the spatio-temporal domain using statistical learning models. The basic approach is clustering procedures for extracting hidden states of the typhoon, and we also analyze the temporal dynamics of the typhoon in terms of transitions between hidden states. The clustering procedures include both spatial and spatio-temporal clustering procedures, including K-means clustering, Self-Organizing Maps (SOM), Mixture of Gaussians (MoG) and Generative Topographic Mapping (GTM) combined with Hidden Markov Model (HMM). The result of clustering is visualized on the ”Evolution Map” on which we analyze and visualize the temporal structure of the typhoon cloud patterns. The results show that spatio-temporal clustering procedures outperform spatial clustering procedures in capturing the temporal structures of the evolution of the typhoon.
KeywordsHide Markov Model Hide State Cluster Procedure Modeling State Transition Generative Topographic Mapping
Unable to display preview. Download preview PDF.
- 3.T. Hastie, R. Tibshirani, and J. Friedman. The Elements of Statistical Learning. Springer, 2001.Google Scholar
- 5.T. Kohonen. Self-Organizing Maps. Springer, second edition, 1997.Google Scholar
- 7.C.M. Bishop, G.E. Hinton, and I.G.D. Strachen. GTM through time. Technical Report NCRG/97/005, Neural Computing Research Group, Aston University, 1997.Google Scholar
- 9.M. Evans, N. Hastings, and B. Peacock. Statistical Distributions. John Wiley & Sons, Inc., third edition, 2000.Google Scholar