Research on Spatial Downscale Temperature Prediction by using Machine Learning and its Application in Urban Heat Island

学生名:王 鋭

研究テーマ:English

(日本語)

入学年月:2018.10

修了年月:2021.09

取得学位:博士（工学）

論文概要:The intensification of urban heat islands (UHIs) due to rapid urbanization poses significant environmental and public health challenges. This research develops a novel spatial downscaling method for air temperature prediction using machine learning algorithms, enabling high-resolution analysis of urban heat islands in five major Japanese metropolitan areas. By addressing the limitations of current atmospheric urban heat island (AUHI) studies, which often rely on low-resolution air temperature data, this research provides a robust methodology for analyzing urban climate dynamics.

The study employs the Extra Trees model to downscale air temperature from a resolution of 1 km to 250 m, using urban structure data and digital elevation models as key predictors. Validation against meteorological data demonstrates high accuracy, confirming the model's reliability for urban climate studies. The research introduces urban heat island intensity (UHII) and urban heat island ratio index (URI) as comparative metrics to evaluate AUHIs across metropolitan areas, highlighting variations in intensity and distribution during winter and summer.

Additionally, a comparative analysis of atmospheric and surface UHIs reveals critical differences in their formation mechanisms and spatial patterns. The study provides actionable insights for urban planning, emphasizing the role of urban structure in mitigating UHI effects. This research advances the understanding of urban climate phenomena and offers practical tools for sustainable urban development.