Scholarlink

<p>Most Urban Heat Island (UHI) research centers on megacities, overlooking coastal and mountainous cities where topography and urban form interact in complex ways. This study focuses on Busan, South Korea, using Local Climate Zones (LCZs), Landsat-derived Land Surface Temperature (LST), and 14 urban morphological and topographic features to explore LST dynamics from 2014 to 2023. Using an XGBoost model (R² = 0.738) with SHAP interpretation, we assessed how each variable influences LST across LCZs. Results show that compact built-up zone LCZ 1 (Compact High-Rise) a) expanded by 330% over the last ten years, while vegetated zones such as LCZ B (Scattered Trees) and LCZ D (Low Plants) declined by 44% and 35%. LCZ 10 (Heavy Industry) had the highest mean LST (21.16&#xa0;°C), while LCZ A (Dense Trees) remained coolest (17.9&#xa0;°C). SHapley Additive exPlanations (SHAP) analysis revealed elevation and distance to water bodies as consistent cooling contributors with mean SHAP values of − 0.45 and − 0.38, while ISF and BCR had strong warming effects (+ 0.42 and + 0.37), especially in LCZs 2, 3, 6, and 10. SVF and FAR exhibited moderate, zone-dependent cooling roles, notably in LCZs 1, 4, and 8. These results highlight that the morphology–LST relationship is highly LCZ-specific and non-linear. By combining LCZ mapping, remote sensing, and explainable machine learning, this study presents a robust framework for understanding UHI formation in topographically complex urban areas, offering targeted insights for thermally adaptive urban planning.</p> Graphical Abstract <p>The graphical abstract in this study presents two fundamental inquiries: (1) how LST and LCZs have evolved over the past decade in Busan (2014–2023), and (2) whether LCZs in Busan influence the effect of urban morphology on LST. These questions shape the study’s analytical framework, which integrates GIS, remote sensing, spatial statistics, and machine learning models to assess urban thermal patterns. As shown in the image, the first section presents the research questions, followed by the methodology. The methodology is structured into three key components: Data Collection, Data Processing, and Data Analysis. Data Collection outlines the types of data gathered, including building morphological data, satellite images, and training samples. Data Processing involves transforming raw data into final datasets, such as LST, LCZ maps, and morphological data, which serve as the foundation for addressing the first research question. Data Analysis applies machine learning models and explainable AI, including XGBoost and SHAP. This step identifies which variables are significant and in which LCZs. Following the methodology, the graphical abstract presents the key findings. The key findings, positioned at the bottom, directly answer the research questions. Overall, the graphical abstract effectively summarizes the study’s focus, providing a clear visual representation of the key analytical processes and findings.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Refining Urban Heat Island Analysis Across Local Climate Zones: Machine Learning Insights on Urban Morphology in Busan

  • 期刊论文

  • 发表日期：

    2025年9月3日

查看全文
  • Rey Jalbuena,
  • Jurng-Jae Yee

摘要

Most Urban Heat Island (UHI) research centers on megacities, overlooking coastal and mountainous cities where topography and urban form interact in complex ways. This study focuses on Busan, South Korea, using Local Climate Zones (LCZs), Landsat-derived Land Surface Temperature (LST), and 14 urban morphological and topographic features to explore LST dynamics from 2014 to 2023. Using an XGBoost model (R² = 0.738) with SHAP interpretation, we assessed how each variable influences LST across LCZs. Results show that compact built-up zone LCZ 1 (Compact High-Rise) a) expanded by 330% over the last ten years, while vegetated zones such as LCZ B (Scattered Trees) and LCZ D (Low Plants) declined by 44% and 35%. LCZ 10 (Heavy Industry) had the highest mean LST (21.16 °C), while LCZ A (Dense Trees) remained coolest (17.9 °C). SHapley Additive exPlanations (SHAP) analysis revealed elevation and distance to water bodies as consistent cooling contributors with mean SHAP values of − 0.45 and − 0.38, while ISF and BCR had strong warming effects (+ 0.42 and + 0.37), especially in LCZs 2, 3, 6, and 10. SVF and FAR exhibited moderate, zone-dependent cooling roles, notably in LCZs 1, 4, and 8. These results highlight that the morphology–LST relationship is highly LCZ-specific and non-linear. By combining LCZ mapping, remote sensing, and explainable machine learning, this study presents a robust framework for understanding UHI formation in topographically complex urban areas, offering targeted insights for thermally adaptive urban planning.

Graphical Abstract

The graphical abstract in this study presents two fundamental inquiries: (1) how LST and LCZs have evolved over the past decade in Busan (2014–2023), and (2) whether LCZs in Busan influence the effect of urban morphology on LST. These questions shape the study’s analytical framework, which integrates GIS, remote sensing, spatial statistics, and machine learning models to assess urban thermal patterns. As shown in the image, the first section presents the research questions, followed by the methodology. The methodology is structured into three key components: Data Collection, Data Processing, and Data Analysis. Data Collection outlines the types of data gathered, including building morphological data, satellite images, and training samples. Data Processing involves transforming raw data into final datasets, such as LST, LCZ maps, and morphological data, which serve as the foundation for addressing the first research question. Data Analysis applies machine learning models and explainable AI, including XGBoost and SHAP. This step identifies which variables are significant and in which LCZs. Following the methodology, the graphical abstract presents the key findings. The key findings, positioned at the bottom, directly answer the research questions. Overall, the graphical abstract effectively summarizes the study’s focus, providing a clear visual representation of the key analytical processes and findings.