Scholarlink

<p>Urban heat island (UHI) effects have become a critical challenge constraining urban sustainability, as accelerating global urbanization has intensified thermal stress in cities worldwide. However, existing studies predominantly rely on correlation analysis and linear regression to explore relationships between urban spatial features and land surface temperature (LST). While recent machine learning studies have improved predictive accuracy, they lack interpretable quantitative methods to systematically identify the marginal benefit boundaries of key driving factors for urban heat environment regulation, thereby hindering the translation of model outputs into actionable, data-driven planning frameworks.Taking Shijiazhuang, a typical high-density city in northern China, as the study area, this study integrates multi-source heterogeneous data including remote sensing imagery, street-view-derived sky view factor (SVF), traffic flow and points of interest, to construct a random forest model combined with explainable AI (XAI) techniques. An XAI-driven “data—model—threshold—decision” framework is developed to characterise nonlinear predictive relationships between urban elements and summer LST via three core quantitative indicators.Results demonstrate that SVF is the most temporally stable LST predictor; key predictive thresholds of 0.270 for building density and 2,800 vehicles/hour for traffic flow are identified; enhanced vegetation index and building morphology exhibit diurnal complementary patterns, with socioeconomic variables showing pronounced nonlinear co-variation with the thermal environment. These findings provide quantitative references for evidence-based, climate-adaptive urban planning.</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

Identifying predictive thresholds and marginal benefit boundaries of urban heat environments via explainable AI

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月22日

查看全文
  • Qiuning Ding,
  • Weizheng Ye,
  • Zichun Shao,
  • Haitao Wang

摘要

Urban heat island (UHI) effects have become a critical challenge constraining urban sustainability, as accelerating global urbanization has intensified thermal stress in cities worldwide. However, existing studies predominantly rely on correlation analysis and linear regression to explore relationships between urban spatial features and land surface temperature (LST). While recent machine learning studies have improved predictive accuracy, they lack interpretable quantitative methods to systematically identify the marginal benefit boundaries of key driving factors for urban heat environment regulation, thereby hindering the translation of model outputs into actionable, data-driven planning frameworks.Taking Shijiazhuang, a typical high-density city in northern China, as the study area, this study integrates multi-source heterogeneous data including remote sensing imagery, street-view-derived sky view factor (SVF), traffic flow and points of interest, to construct a random forest model combined with explainable AI (XAI) techniques. An XAI-driven “data—model—threshold—decision” framework is developed to characterise nonlinear predictive relationships between urban elements and summer LST via three core quantitative indicators.Results demonstrate that SVF is the most temporally stable LST predictor; key predictive thresholds of 0.270 for building density and 2,800 vehicles/hour for traffic flow are identified; enhanced vegetation index and building morphology exhibit diurnal complementary patterns, with socioeconomic variables showing pronounced nonlinear co-variation with the thermal environment. These findings provide quantitative references for evidence-based, climate-adaptive urban planning.