HR: 0800h
AN: H31B-0357    [Abstracts]
TI: Probabilistic Prediction Of Heavy Rainfall Using Pattern Recognition Technique Based On Self-Organizing Map (SOM)
AU: * NISHIYAMA, K
EM: nisiyama@civil.kyushu-u.ac.jp
AF: Fuculty of Engineering, Kyushu University, 744, Motooka, Nishi-ku, Fukuoka, 8190395, Japan
AU: JINNO, K
EM: jinno@civil.kyushu-u.ac.jp
AF: Fuculty of Engineering, Kyushu University, 744, Motooka, Nishi-ku, Fukuoka, 8190395, Japan
AU: WAKIMIZU, K
EM: wakimizu@bpes.kyushu-u.ac.jp
AF: Fuculty of Agriculture, Kyushu University, 6-10-1, Hakozaki, Higashi-ku, Fukuoka, 8128581, Japan
AB: Most of the heavy rainfall systems are closely related to spatially-extended meteorological information, in other words, multi-dimensional information. Therefore, in this study, pattern recognition technique based on Self- Organizing Map (SOM) was applied to the prediction of heavy rainfall in the rainy season in Japan in combination with Back Propagation (BP: supervised ANN). The SOM is an unsupervised ANN-based pattern recognition technique, projecting high-dimensional input variables onto two-dimensional regularly-arranged units for visualization. Here, the patterns of meteorological field characterizing the rainy season (BAIU) in Japan were classified using the SOM, and related to rainfall data using the BP. From the results, the rainfall prediction technique succeeded in constructing meteorologically-significant relationships between complicated meteorological field patterns and rainfall by focusing on the inherent properties of the SOM. Particularly, it was clearly shown that high probability of heavy rainfall corresponds to a meteorological field pattern characterized by Low-Level Jet (LLJ) and ample water vapor, which was closely associated with disastrous rainfall events in Japan.
DE: 1817 Extreme events
DE: 1840 Hydrometeorology
DE: 1854 Precipitation (3354)
SC: Hydrology [H]
MN: 2007 Fall Meeting