Behavior of Cool and Hot/Warm Air Plumes in Tokyo Revealed by Artificial Intelligence

The primary objective of this study was to visualize cool air plumes generated in and coming from urban green areas. Differences in air temperature at meteorological stations at the edges of urban green areas and an urban site in Tokyo in August, November, January, and April were acquired at a time resolution of 10 min for 3 or 6 years. By handling more than 12,800 cases, the automatic neuro-evolution of a deep learning architecture was applied to generalize and visualize the typical wind directions and speeds when cool air plumes were in effect at meteorological stations at the edges of the urban green areas. In some cases, the visualization of cool air plume was successful. For instance, the Tokyo site at the edge of a green area was typically cooler than the urban site in Itabashi when the Tokyo site was subjected to calm wind (< 3 m s⁻¹) from the urban green area at night in August. However, in some cases in August, the Tokyo site was hotter than the Itabashi site when the Tokyo site had strong winds from every direction in the daytime, indicating that the strong winds brought heat from heat sources surrounding the urban green area. Thus, the cooling effect at the Tokyo site was at microscale or local scale, while the heating/warming effect was mesoscale. This emphasized that, especially in the daytime in summer, the cool air plumes were prone to be engulfed by greater volumes of hot air plumes from various heat sources. Mechanisms of other related findings are further discussed.