Statistical analysis on heavy rainfall areas causing serious disasters in Kyushu, Japan, using Self-Organizing Map

The Japanese islands are highly vulnerable to heavy rainfall during warm periods due to the inflow of warm, moist air, which often results in disasters such as floods and landslides. In particular, the western region, Kyushu, experiences frequent heavy rainfall, including linear rainbands that can persist for several hours. This study aims to investigate the characteristics of heavy rainfall areas in Kyushu using a self-organizing map (SOM), a type of unsupervised neural network and pattern recognition method.

Initially, we applied the SOM algorithm to train average meteorological fields over three-hour intervals, consisting of 850 hPa winds and equivalent potential temperature from 1979 to 2023, including the Kyushu region. The patterns obtained were then compared to heavy rainfall areas from 2006 to 2023, defined as regions with 3-hour rainfall amounts exceeding 100 mm/3h over an area of at least 500 km², with a maximum value of at least 150 mm/3h within that region.

To extract these heavy rainfall areas, we identified the closed regions with precipitation exceeding 100 mm/3h from the actual rainfall distribution. Principal component analysis (PCA) was then applied to the coordinates of the heavy rainfall areas, and the first principal component, which maximizes the variance, was considered the axis of movement for the heavy rainfall area. The first and second principal components correspond to the major and minor axes of the heavy rainfall area, respectively. Finally, we placed a rectangle parallel to the major axis, adjusting its four sides to touch the edges of the heavy rainfall area, which helped determine its shape.

The analysis revealed that heavy rainfall areas are more frequent in meteorological fields associated with the inflow of warm, moist air from the southwest or west-southwest into the Kyushu region. Approximately 70% of these areas were identified as elongated precipitation bands with an axis ratio greater than 2.5. Additionally, heavy rainfall areas tend to occur most frequently between 3:00 and 9:00 AM. Furthermore, the occurrence of heavy rainfall areas was particularly high in meteorological fields associated with fronts. These areas were concentrated on the western side of Kyushu, with very little occurrence on the eastern side. The probability of occurrence was about 10% during the daytime but exceeded 20% between 6:00 and 9:00 AM.

Based on these findings, it is crucial to recognize that heavy rainfall areas in Kyushu are especially prominent at night. In particular, when the region is covered by meteorological fields associated with fronts and warm, moist air, with an equivalent potential temperature exceeding 340 K, heavy rainfall during nighttime hours should be closely monitored. These insights should be integrated into the disaster prevention plans of local governments.