Development of a Japanese version of Local Climate Zone using machine learning

The Local Climate Zone (LCZ) classification has been widely utilized not only in urban climate studies but also in various research fields. However, previous studies have pointed out that the classification of LCZ classes may not be appropriate for certain urban areas, including those in Japan. Therefore, this study aims to develop a "Japanese Version of LCZ" that is better suited to Japan's urban environment by redefining LCZ classes using high-quality urban block and natural geographic data, such as the PLATEAU dataset provided by the Ministry of Land, Infrastructure, Transport and Tourism. Osaka City (Osaka Prefecture), Shizuoka City (Shizuoka Prefecture), and Saitama City (Saitama Prefecture) were selected as study areas. The classification parameters included building coverage ratio, average building height, impervious surface ratio, surface heat transfer coefficient, number of buildings, number of building floors, and energy consumption. Each parameter was calculated for a 100 m mesh within the study areas (N = 68,681). Classification was performed by first applying standardized parameters to a Self-Organizing Map (SOM) and then clustering using the k-means method. The PLATEAU dataset was used to generate the parameters. Furthermore, to investigate the relationship between the developed Japanese LCZ classes and meteorological conditions, heat stress observations were conducted in Osaka’s urban area on September 4-5, 2024. The classification results indicate that each cluster was appropriately categorized based on the characteristics of Japanese urban morphology and land cover. As a result, a total of 11 urban-type Japanese LCZ classes were proposed. In particular, a new category was defined for residential areas with a high concentration of wooden structures. These findings suggest that the Japanese LCZ more accurately represents Japan’s urban environments compared to the conventional LCZ classification.