Do Different Urban Morphologies Have Consistent Patterns on Urban Heat Islands in Different Cities and Scales? A Local Climate Zone Exploration

Abstract

Rapid urbanization exacerbates the urban heat island (UHI) effect, posing challenges to the United Nations' Sustainable Development Goals amidst global warming. Attributing UHI impacts to urban morphology is increasingly crucial, yet few studies incorporate local climate zone (LCZ) in this context, especially when combined with the interpretability of machine learning in diverse cities and building structures, leading to unclear rule and conclusion among them. This study integrates various dimensional indicators based on multisource remote sensing data with land use and land cover and employs an integrated machine learning model to predict UHI in Beijing and Tianjin, generalizing the model to Shijiazhuang. The Shapley additive explanations method is used to analyze the influence of each feature at global, local, and LCZ scales. Results indicate that extreme gradient boosting achieves the highest accuracy compared to other models, with its accuracy being improved by 10% or more; compared with general land classification, LCZ-based classification has more than doubled the validation accuracy, smoothing sharp contrasts while preserving details. The factors affecting UHI in different cities and architectural form have partial consistency, with population growth and vegetation reduction being identified as major factors contributing to significant UHI increases. Our findings provide scientific references for understanding UHI drivers across multiple scales and regions and optimizing urban morphology to mitigate urban thermal environments.

Keywords

• Local climate zone (LCZ)
• machine learning
• remote sensing
• Shapley additive explanations (SHAP) method
• urban heat island (UHI)
• urban morphology