Interpretation of typhoon cloud patterns by holistic analysis

摘要

The main theme of this paper is the morphological analysis of typhoons. That is, we aim at establishing a new scheme for estimating the intensity of typhoons based on the morphological analysis of typhoons. The infrastructure of this research is our typhoon image collection that consists of about 20,000 typhoon images in the northern hemisphere, and about 9,000 typhoon images in the southern hemisphere. Based on this testbed, we applied several holistic analysis approaches such as principal component analysis for the challenge of typhoon analysis. As a result, the eigenpictures of the typhoons well represent the characteristics of typhoon cloud patterns such as eyewall clouds and rain bands; moreover, the application of multiscale analysis to those eigenpictures yields further information related to the scale of meteorological phenomena. However, although holistic analysis shows good performance in similarity-based image retrieval, typhoon prediction based on similarity-based image retrieval is found to be a difficult task due to the chaotic nature of the atmosphere.