Our research aims at discovering useful knowledge from the large collection of satellite images of typhoons using data mining approaches. In addition, this paper discusses more ambitious challenges, namely the discovery of the signs of infrequent but relevant events from typhoon cloud patterns observed from satellites. For this purpose, we have built the typhoon image collection of around 34,000 images for the northern and southern hemisphere, which provides medium-sized, richly-variational and quality-controlled data archive suitable for data mining research. Our preferred approach to typhoon data mining is oriented toward a probability and statistical framework, and our goal is to learn the characteristics of the space of typhoon cloud patterns from satellite images. We introduce the PCA-based representation of typhoons, and its application to the discovery of the signs of the rapid intensification of the typhoon. We also present meteorological background of the issues, comparison of approaches to those issues, and promising areas of chance discovery research for the typhoon.

      @InProceedings{ k:fai02,
      author = {Asanobu KITAMOTO},
      title = {Discovery of the Signs in Typhoon Image Collection},
      booktitle = {Technical Report of Japanese Society for Artificial Intelligence (JSAI)},
      volume = {SIG-FAI-A103},
      pages = {19-26},
      year = 2002,
      month = 1,
      doi = {},
      note = { (in Japanese)},
      }