Frequency analysis of heavy rainfall related to synoptic weather patterns in Kyushu Island, Japan by using the Self-Organizing Map (SOM)

Abstract: Recently heavy rainfall frequency in Kyushu Island, the southwestern part of Japan, during the rainy season, has been increasing due to possibly climate change, global warming. Owing to the recent increase of heavy rainfall frequency, natural disasters related to heavy rainfall including flooding and landslides have occurred and caused serious damages to infrastructures and human lives. Therefore, for recognizing the possibility and the risk of heavy rainfall related natural disasters, it is very important to identify and investigate that what kinds of meteorological fields have frequently caused heavy rainfall so far and significantly contributed to the recent increase of heavy rainfall frequency, specifically in Kyushu Island. Hence, the objective of this study is to reveal the aforementioned meteorological fields by correlating annual variations in heavy rainfall frequency measured for the study period of 39 years with synoptic weather groups classified by the Self-Organizing Map (SOM) developed by Kohonen (1995) and subsequent two clustering techniques (Ward method and K-means method). The SOM algorithm is an unsupervised neural network algorithm that can non-linearly classify complex high-dimensional input data into visually-understanding patterns on a two dimensional map. In this study, 19032 meteorological fields recorded for 39 years were classified into 45 synoptic weather groups depending on the similarity. Subsequently, annual variations in heavy rainfall frequency corresponding to 45 synoptic weather groups have been investigated. The result shows that stationary front, typhoon, a combination of stationary front and low pressure, and moist tongue have frequently caused heavy rainfall in Kyushu Island so far. Moreover, these meteorological fields have highly contributed to the recent increase of heavy rainfall frequency. Therefore, when those meteorological fields appear in Kyushu Island, it is considerably important to carefully check weather conditions for recognizing the possibility and risk of natural disasters associated to heavy rainfall. Note that whereas heavy rainfall frequency has increased recently in those meteorological fields, the increase trend of appearance frequencies of those fields cannot be recognized because the occurrence of heavy rainfall is often controlled by meteorological factors such as regional topographical features. Therefore, further investigations on smaller-scales will be needed in the future work.